Neural network prediction of obstructive sleep apnea from clinical criteria

Artificial intelligence algorithms permits rapid acute kidney injury risk classification of patients with acute myocardial infarction

Objective

This study aimed to develop and validate several artificial intelligence (AI) models to identify acute myocardial infarction (AMI) patients at an increased risk of acute kidney injury (AKI) during hospitalization.

Methods

Included were patients diagnosed with AMI from the Medical Information Mart for Intensive Care (MIMIC) III and IV databases. Two cohorts of AMI patients from Changzhou Second People's Hospital and Xuzhou Center Hospital were used for external validation of the models. Patients' demographics, vital signs, clinical characteristics, laboratory results, and therapeutic measures were extracted. Totally, 12 AI models were developed. The area under the receiver operating characteristic curve (AUC) were calculated and compared.

Results

AKI occurred during hospitalization in 1098 (28.3 %) of the 3882 final enrolled patients, split into training (3105) and test (777) sets randomly. Among them, Random Forest (RF), C5.0 and Bagged CART models outperformed the other models in both the training and test sets. The AUCs for the test set were 0.754, 0.734 and 0.730, respectively. The incidence of AKI was 9.8 % and 9.5 % in 2202 patients in the Changzhou cohort and 807 patients in the Xuzhou cohort with AMI, respectively. The AUCs for patients in the Changzhou cohort were RF, 0.761; C5.0, 0.733; and bagged CART, 0.725, respectively, and Xuzhou cohort were RF, 0.799; C5.0, 0.808; and bagged CART, 0.784, respectively.

Conclusion

Several machines learning-based prediction models for AKI after AMI were developed and validated. The RF, C5.0 and Bagged CART model performed robustly in identifying high-risk patients earlier.

Clinical trial approval statement

This Trial was registered in the Chinese clinical trials registry: ChiCTR1800014583. Registered January 22, 2018 (http://www.chictr.org.cn/searchproj.aspx).

Application of artificial intelligence in the diagnosis of sleep apnea

STUDY OBJECTIVES

Machine learning (ML) models have been employed in the setting of sleep disorders. This review aims to summarize the existing data about the role of ML techniques in the diagnosis, classification, and treatment of sleep-related breathing disorders.

METHODS

A systematic search in Medline, EMBASE, and Cochrane databases through January 2022 was performed.

RESULTS

Our search strategy revealed 132 studies that were included in the systematic review. Existing data show that ML models have been successfully used for diagnostic purposes. Specifically, ML models showed good performance in diagnosing sleep apnea using easily obtained features from the electrocardiogram, pulse oximetry, and sound signals. Similarly, ML showed good performance for the classification of sleep apnea into obstructive and central categories, as well as predicting apnea severity. Existing data show promising results for the ML-based guided treatment of sleep apnea. Specifically, the prediction of outcomes following surgical treatment and optimization of continuous positive airway pressure therapy can be guided by ML models.

CONCLUSIONS

The adoption and implementation of ML in the field of sleep-related breathing disorders is promising. Advancements in wearable sensor technology and ML models can help clinicians predict, diagnose, and classify sleep apnea more accurately and efficiently.

CITATION

Bazoukis G, Bollepalli SC, Chung CT, et al. Application of artificial intelligence in the diagnosis of sleep apnea. J Clin Sleep Med. 2023;19(7):1337-1363.

Review of Application of Machine Learning as a Screening Tool for Diagnosis of Obstructive Sleep Apnea

Obstructive sleep apnea syndrome (OSAS) is a pervasive disorder with an incidence estimated at 5-14 percent among adults aged 30-70 years. It carries significant morbidity and mortality risk from cardiovascular disease, including ischemic heart disease, atrial fibrillation, and cerebrovascular disease, and risks related to excessive daytime sleepiness. The gold standard for diagnosis of OSAS is the polysomnography (PSG) test which requires overnight evaluation in a sleep laboratory and expensive infrastructure, which renders it unsuitable for mass screening and diagnosis. Alternatives such as home sleep testing need patients to wear diagnostic instruments overnight, but accuracy continues to be suboptimal while access continues to be a barrier for many. Hence, there is a continued significant underdiagnosis and under-recognition of sleep apnea in the community, with at least one study suggesting that 80-90% of middle-aged adults with moderate to severe sleep apnea remain undiagnosed. Recently, we have seen a surge in applications of artificial intelligence and neural networks in healthcare diagnostics. Several studies have attempted to examine its application in the diagnosis of OSAS. Signals included in data analytics include Electrocardiogram (ECG), photo-pletysmography (PPG), peripheral oxygen saturation (SpO2), and audio signals. A different approach is to study the application of machine learning to use demographic and standard clinical variables and physical findings to try and synthesize predictive models with high accuracy in assisting in the triage of high-risk patients for sleep testing. The current paper will review this latter approach and identify knowledge gaps that may serve as potential avenues for future research.

The Predictive Role of Subcutaneous Adipose Tissue in the Pathogenesis of Obstructive Sleep Apnoea

Simple Summary

Although several methods are used to diagnose obstructive sleep apnoea (OSA), the disorder is still underdiagnosed, leading to public healthcare problems. The main aim of the present study was to analyse the role of artificial intelligence in OSA diagnostics and obstruction localisation and, moreover, the role of subcutaneous adipose tissue in OSA pathophysiology. The significance of the present investigation is that using US in OSA diagnostics and obstruction location, an additional opportunity besides standard procedures (i.e., drug-induced sleep endoscopy or polygraphy) is presented, which is vital due to the high number of undiagnosed cases. Applying the algorithm, including artificial intelligence, the presence of obstructions and its localisation, can be determined with high precision. This can be essential in therapy planning or preoperative patient preparation.

Abstract

Introduction: Our aim was to investigate the applicability of artificial intelligence in predicting obstructive sleep apnoea (OSA) and upper airway obstruction using ultrasound (US) measurements of subcutaneous adipose tissues (SAT) in the regions of the neck, chest and abdomen. Methods: One hundred patients were divided into mild (32), moderately severe-severe (32) OSA and non-OSA (36), according to the results of the polysomnography. These patients were examined using anthropometric measurements and US of SAT and drug-induced sleep endoscopy. Results: Using SAT US and anthropometric parameters, oropharyngeal obstruction could be predicted in 64% and tongue-based obstruction in 72%. In predicting oropharyngeal obstruction, BMI, abdominal and hip circumferences, submental SAT and SAT above the second intercostal space on the left were identified as essential parameters. Furthermore, tongue-based obstruction was predicted mainly by height, SAT measured 2 cm above the umbilicus and submental SAT. The OSA prediction was successful in 97% using the parameters mentioned above. Moreover, other parameters, such as US-based SAT, with SAT measured 2 cm above the umbilicus and both-sided SAT above the second intercostal spaces as the most important ones. Discussion: Based on our results, several categories of OSA can be predicted using artificial intelligence with high precision by using SAT and anthropometric parameters.

Predictive performances of 6 data mining techniques for obstructive sleep apnea-hypopnea syndrome

PURPOSE

This study compared the effects of 6 types of obstructive sleep apnea-hypopnea syndrome (OSAHS) prediction models to develop a reference for selecting OSAHS data mining tools in clinical practice.

METHODS

This cross-sectional study included 401 cases. They were randomly divided into 2 groups: training (70%) and testing (30%). Logistic regression, a Bayesian network, an artificial neural network, a support vector learning machine, C5.0, and a classification and regression tree were each adopted to establish 6 prediction models. After training, the 6 models were used to test the remaining samples and calculate the correct and error rates of each model.

RESULTS

Twenty-one input variables for which the difference between the patient and nonpatient groups was statistically significant were considered. The models found the abdominal circumference, neck circumference, and nocturia ≥2 per night to be the most important variables. The support vector machine, neural network, and C5.0 models performed better than the classification and regression tree, Bayesian network, and logistic regression models.

CONCLUSIONS

In terms of predicting the risk of OSAHS, the support vector machine, neural network, and C5.0 were superior to the classification and regression tree, Bayesian network, and logistic regression models. However, such results were obtained based on the data of a single center, so they need to be further validated by other institutions.

Optimizing acute stroke outcome prediction models: Comparison of generalized regression neural networks and logistic regressions

Background

Generalized regression neural network (GRNN) and logistic regression (LR) are extensively used in the medical field; however, the better model for predicting stroke outcome has not been established. The primary goal of this study was to compare the accuracies of GRNN and LR models to identify the most optimal model for the prediction of acute stroke outcome, as well as explore useful biomarkers for predicting the prognosis of acute stroke patients.

Method

In a single-center study, 216 (80% for the training set and 20% for the test set) acute stroke patients admitted to the Shenzhen Second People’s Hospital between December 2019 to June 2021 were retrospectively recruited. The functional outcomes of the patients were measured using Barthel Index (BI) on discharge. A training set was used to optimize the GRNN and LR models. The test set was utilized to validate and compare the performances of GRNN and LR in predicting acute stroke outcome based on the area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, and the Kappa value.

Result

The LR analysis showed that age, the National Institute Health Stroke Scale score, BI index, hemoglobin, and albumin were independently associated with stroke outcome. After validating in test set using these variables, we found that the GRNN model showed a better performance based on AUROC (0.931 vs 0.702), sensitivity (0.933 vs 0.700), specificity (0.889 vs 0.722), accuracy (0.896 vs 0.729), and the Kappa value (0.775 vs 0.416) than the LR model.

Conclusion

Overall, the GRNN model demonstrated superior performance to the LR model in predicting the prognosis of acute stroke patients. In addition to its advantage in not affected by implicit interactions and complex relationship in the data. Thus, we suggested that GRNN could be served as the optimal statistical model for acute stroke outcome prediction. Simultaneously, prospective validation based on more variables of the GRNN model for the prediction is required in future studies.

Helping early obstructive sleep apnea diagnosis with machine learning: A systematic review (Preprint)

Background

American Academy of Sleep Medicine guidelines suggest that clinical prediction algorithms can be used to screen patients with obstructive sleep apnea (OSA) without replacing polysomnography, the gold standard.

Objective

We aimed to identify, gather, and analyze existing machine learning approaches that are being used for disease screening in adult patients with suspected OSA.

Methods

We searched the MEDLINE, Scopus, and ISI Web of Knowledge databases to evaluate the validity of different machine learning techniques, with polysomnography as the gold standard outcome measure and used the Prediction Model Risk of Bias Assessment Tool (Kleijnen Systematic Reviews Ltd) to assess risk of bias and applicability of each included study.

Results

Our search retrieved 5479 articles, of which 63 (1.15%) articles were included. We found 23 studies performing diagnostic model development alone, 26 with added internal validation, and 14 applying the clinical prediction algorithm to an independent sample (although not all reporting the most common discrimination metrics, sensitivity or specificity). Logistic regression was applied in 35 studies, linear regression in 16, support vector machine in 9, neural networks in 8, decision trees in 6, and Bayesian networks in 4. Random forest, discriminant analysis, classification and regression tree, and nomogram were each performed in 2 studies, whereas Pearson correlation, adaptive neuro-fuzzy inference system, artificial immune recognition system, genetic algorithm, supersparse linear integer models, and k-nearest neighbors algorithm were each performed in 1 study. The best area under the receiver operating curve was 0.98 (0.96-0.99) for age, waist circumference, Epworth Somnolence Scale score, and oxygen saturation as predictors in a logistic regression.

Conclusions

Although high values were obtained, they still lacked external validation results in large cohorts and a standard OSA criteria definition.

Trial Registration

PROSPERO CRD42021221339; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=221339

Privacy-preserving federated neural network learning for disease-associated cell classification

Training accurate and robust machine learning models requires a large amount of data that is usually scattered across data silos. Sharing or centralizing the data of different healthcare institutions is, however, unfeasible or prohibitively difficult due to privacy regulations. In this work, we address this problem by using a privacy-preserving federated learning-based approach, PriCell, for complex models such as convolutional neural networks. PriCell relies on multiparty homomorphic encryption and enables the collaborative training of encrypted neural networks with multiple healthcare institutions. We preserve the confidentiality of each institutions’ input data, of any intermediate values, and of the trained model parameters. We efficiently replicate the training of a published state-of-the-art convolutional neural network architecture in a decentralized and privacy-preserving manner. Our solution achieves an accuracy comparable with the one obtained with the centralized non-secure solution. PriCell guarantees patient privacy and ensures data utility for efficient multi-center studies involving complex healthcare data.

•

We enable collaborative and privacy-preserving model training between institutions

•

Training under encryption does not degrade the utility of the data

•

We apply our solution to the single-cell analysis in a federated setting

•

Our method is generalizable to other machine learning tasks in the healthcare domain

High-quality medical machine learning models will benefit greatly from collaboration between health care institutions. Yet, it is usually difficult to transfer data between these institutions due to strict privacy regulations. In this study, we propose a solution, PriCell, that relies on multiparty homomorphic encryption to enable privacy-preserving collaborative machine learning while protecting via encryption the institutions' input data, the model, and any value exchanged between the institutions. We show the maturity of our solution by training a published state-of-the-art convolutional neural network in a decentralized and privacy-preserving manner. We compare the accuracy achieved by PriCell with the centralized and non-secure solutions and show that PriCell guarantees privacy without reducing the utility of the data. The benefits of PriCell constitute an important landmark for real-world applications of collaborative training while preserving privacy.

Barriers of artificial intelligence implementation in the diagnosis of obstructive sleep apnea

Background

Obstructive sleep apnea is a common clinical condition and has a significant impact on the health of patients if untreated. The current diagnostic gold standard for obstructive sleep apnea is polysomnography, which is labor intensive, requires specialists to utilize, expensive, and has accessibility challenges. There are also challenges with awareness and identification of obstructive sleep apnea in the primary care setting. Artificial intelligence systems offer the opportunity for a new diagnostic approach that addresses the limitations of polysomnography and ultimately benefits patients by streamlining the diagnostic expedition.

Main body

The purpose of this project is to elucidate the barriers that exist in the implementation of artificial intelligence systems into the diagnostic context of obstructive sleep apnea. It is essential to understand these challenges in order to proactively create solutions and establish an efficient adoption of this new technology. The literature regarding the evolution of the diagnosis of obstructive sleep apnea, the role of artificial intelligence in the diagnosis, and the barriers in artificial intelligence implementation was reviewed and analyzed.

Conclusion

The barriers identified were categorized into different themes including technology, data, regulation, human resources, education, and culture. Many of these challenges are ubiquitous across artificial intelligence implementation in any medical diagnostic setting. Future research directions include developing solutions to the barriers presented in this project.

Graphical abstract

Explainable fuzzy neural network with easy-to-obtain physiological features for screening obstructive sleep apnea-hypopnea syndrome

OBJECTIVE/BACKGROUND

Recently, several tools for screening obstructive sleep apnea-hypopnea syndrome (OSAHS) have been devised with varied shortcomings. To overcome these drawbacks, we aimed to propose a self-estimation method using an explainable prediction model with easy-to-obtain variables and evaluate its performance for predicting OSAHS.

PATIENTS/METHODS

This retrospective, cross-sectional study selected significant easy-to-obtain variables from patients, suspected of having OSAHS by regression analysis, and fed these variables into the proposed explainable fuzzy neural network (EFNN), a back propagation neural network (BPNN) and a stepwise regression model to compare the screening performance for OSAHS.

RESULTS

Of the 300 participants, three easily available features, such as waist circumference, mean blood pressure (BP) at the end of polysomnography and the difference in systolic BP between the end and start of polysomnography, were obtained from regression analysis with a five-fold cross-validation scheme. Feeding these three variables into the prediction models showed that the average prediction differences for apnea-hypopnea index (AHI) when using the EFNN, BPNN, and regression model were respectively 1.5 ± 18.2, 3.5 ± 19.1 and 0.1 ± 19.3, indicating none of the tested methods had good efficacy to predict the AHI values. The performance as determined by the sensitivity + specificity-1 value for screening moderate-to-severe OSAHS of the EFNN, BPNN and regression model were respectively 0.440, 0.414 and 0.380.

CONCLUSIONS

When fed with easy-to-obtain physiological features, the understandable EFNN should be the preferred method to predict moderate-to-severe OSAHS.

Support vector machine prediction of obstructive sleep apnea in a large-scale Chinese clinical sample

STUDY OBJECTIVES

Polysomnography is the gold standard for diagnosis of obstructive sleep apnea (OSA) but it is costly and access is often limited. The aim of this study is to develop a clinically useful support vector machine (SVM)-based prediction model to identify patients with high probability of OSA for nonsleep specialist physician in clinical practice.

METHODS

The SVM model was developed using the features routinely collected at the clinical evaluation from 6,875 Chinese patients referred to sleep clinics for suspected OSA. Three apnea-hypopnea index (AHI) cutoffs, ≥5/h, ≥15/h, and ≥30/h were used to define the severity of OSA. The continuous and categorized features were selected separately and were further selected through stepwise forward feature selection. The modeling was achieved through fivefold cross-validation. The model discriminative ability was evaluated for the whole data set and four subgroups categorized with gender and age (<65 versus ≥65 years old [y/o]).

RESULTS

Two features were selected to predict AHI cutoff ≥5/h with six features selected for ≥15/h, and six features selected for ≥30/h, respectively, to reach Area under the Receiver Operating Characteristic (AUROC) 0.82, 0.80, and 0.78, respectively. The sensitivity was 74.14%, 75.18%, and 70.26%, while the specificity was 74.71%, 68.73%, and 70.30%, respectively. Compared to logistic regression, Berlin questionnaire, NoSAS Score, and Supersparse Linear Integer Model (SLIM) scoring system, the SVM model performs better with a more balanced sensitivity and specificity. The discriminative ability was best for male <65 y/o and modest for female ≥65 y/o.

CONCLUSION

Our model provides a simple and accurate modality for early identification of patients with OSA and may potentially help prioritize them for sleep study.

Diagnostic Accuracy of Oxygen Desaturation Index for Sleep-Disordered Breathing in Patients With Diabetes

BACKGROUND

Polysomnography (PSG) is the gold standard for diagnosis of sleep-disordered breathing (SDB). But it is impractical to perform PSG in all patients with diabetes. The objective was to develop a clinically easy-to-use prediction model to diagnosis SDB in patients with diabetes.

METHODS

A total of 440 patients with diabetes were recruited and underwent overnight PSG at West China Hospital. Prediction algorithms were based on oxygen desaturation index (ODI) and other variables, including sex, age, body mass index, Epworth score, mean oxygen saturation, and total sleep time. Two phase approach was employed to derivate and validate the models.

RESULTS

ODI was strongly correlated with apnea-hypopnea index (AHI) (r_s = 0.941). In the derivation phase, the single cutoff model with ODI was selected, with area under the receiver operating characteristic curve (AUC) of 0.956 (95%CI 0.917-0.994), 0.962 (95%CI 0.943-0.981), and 0.976 (95%CI 0.956-0.996) for predicting AHI ≥5/h, ≥15/h, and ≥30/h, respectively. We identified the cutoff of ODI 5/h, 15/h, and 25/h, as having important predictive value for AHI ≥5/h, ≥15/h, and ≥30/h, respectively. In the validation phase, the AUC of ODI was 0.941 (95%CI 0.904-0.978), 0.969 (95%CI 0.969-0.991), and 0.949 (95%CI 0.915-0.983) for predicting AHI ≥5/h, ≥15/h, and ≥30/h, respectively. The sensitivity of ODI ≥5/h, ≥15/h, and ≥25/h was 92%, 90%, and 93%, respectively, while the specificity was 73%, 89%, and 85%, respectively.

CONCLUSIONS

ODI is a sensitive and specific tool to predict SDB in patients with diabetes.

Differentiation of Clinical Patterns Associated With Rhinologic Disease

BACKGROUND

Common rhinologic diagnoses have similar presentations with a varying degree of overlap. Patterns may exist within clinical data that can be useful for early diagnosis and predicting outcomes.

OBJECTIVE

To explore the feasibility of artificial intelligence to differentiate patterns in patient data in order to develop clinically-meaningful diagnostic groups.

METHODS

A cross-sectional study of prospectively-acquired patient data at a tertiary rhinology clinic was performed. Data extracted included objective findings on nasal endoscopy, patient reported quality of life (PRQOL) instrument ratings, peripheral eosinophil fraction, and past medical history. Unsupervised non-hierarchical cluster analysis was performed to discover patterns in the data using 22 input variables.

RESULTS

A total of 545 patients were analyzed after application of inclusion and exclusion criteria yielding 7 unique patient clusters, highly dependent on PRQOL scores and demographics. The clusters were clinically-relevant with distinct characteristics. Chronic rhinosinusitis without nasal polyposis (CRSsNP) was associated with two clusters having low frequencies of asthma and low eosinophil fractions. Chronic rhinosinusitis with nasal polyposis (CRSwNP) was associated with high frequency of asthma, mean (standard deviation [SD]) NOSE scores of 66 (19) and SNOT-22 scores of 41 (15), and high eosinophil fractions. AR was present in multiple clusters. RARS was associated with the youngest population with mean (SD) NOSE score of 54 (23) and SNOT-22 score of 41 (19).

CONCLUSION

Broader consideration of initially available clinical data may improve diagnostic efficiency for rhinologic conditions without ancillary studies, using computer-driven algorithms. PRQOL scores and demographic information appeared to be useful adjuncts, with associations to diagnoses in this pilot study.

Application of machine learning to predict obstructive sleep apnea syndrome severity

Introduction:

Obstructive sleep apnea syndrome has become an important public health concern. Polysomnography is traditionally considered an established and effective diagnostic tool providing information on the severity of obstructive sleep apnea syndrome and the degree of sleep fragmentation. However, the numerous steps in the polysomnography test to diagnose obstructive sleep apnea syndrome are costly and time consuming. This study aimed to test the efficacy and clinical applicability of different machine learning methods based on demographic information and questionnaire data to predict obstructive sleep apnea syndrome severity.

Materials and methods:

We collected data about demographic characteristics, spirometry values, gas exchange (PaO2, PaCO2) and symptoms (Epworth Sleepiness Scale, snoring, etc.) of 313 patients with previous diagnosis of obstructive sleep apnea syndrome. After principal component analysis, we selected 19 variables which were used for further preprocessing and to eventually train seven types of classification models and five types of regression models to evaluate the prediction ability of obstructive sleep apnea syndrome severity, represented either by class or by apnea–hypopnea index. All models are trained with an increasing number of features and the results are validated through stratified 10-fold cross validation.

Results:

Comparative results show the superiority of support vector machine and random forest models for classification, while support vector machine and linear regression are better suited to predict apnea–hypopnea index. Also, a limited number of features are enough to achieve the maximum predictive accuracy. The best average classification accuracy on test sets is 44.7 percent, with the same average sensitivity (recall). In only 5.7 percent of cases, a severe obstructive sleep apnea syndrome (class 4) is misclassified as mild (class 2). Regression results show a minimum achieved root mean squared error of 22.17.

Conclusion:

The problem of predicting apnea–hypopnea index or severity classes for obstructive sleep apnea syndrome is very difficult when using only data collected prior to polysomnography test. The results achieved with the available data suggest the use of machine learning methods as tools for providing patients with a priority level for polysomnography test, but they still cannot be used for automated diagnosis.

A Novel Artificial Neural Network Based Sleep-Disordered Breathing Screening Tool

STUDY OBJECTIVES

This study evaluated a novel artificial neural network (ANN) based sleep-disordered breathing (SDB) screening tool incorporating nocturnal pulse oximetry with demographic, anatomic, and clinical data. The tool was compatible with 6 categories of apnea-hypopnea index (AHI) with 4% oxyhemoglobin desaturation threshold, ≥ 5, 10, 15, 20, 25, and 30 events/h.

METHODS

Using a general population dataset, the training set included 2,280 subjects, whereas the test set included 470 subjects. The input of this tool was a set of 22 variables. The tool had six neural network models for each AHI threshold. Several metrics were explored to evaluate the performance of the tool: area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value, negative predictive value, and 95% confidence interval (CI).

RESULTS

The AUC was 0.904, 0.912, 0.913, 0.926, 0.930, and 0.954, respectively, with models of AHI ≥ 5, 10, 15, 20, 25, and 30 events/h thresholds. The sensitivities of all neural network models were higher than 95%. The AHI ≥ 30 events/h model had the maximum sensitivity: 98.31% (95% CI: 95.01%-100%).

CONCLUSIONS

The results of this study suggested that the ANN based SDB screening tool can be used to identify the presence or absence of SDB. Future validation should be performed in other populations to determine the practicability of this screening tool in sleep clinics and other at-risk populations.

Can Statistical Machine Learning Algorithms Help for Classification of Obstructive Sleep Apnea Severity to Optimal Utilization of Polysomno graphy Resources?

Objectives: The goal of this study is to evaluate the results of machine learning methods for the classification of OSA severity of patients with suspected sleep disorder breathing as normal, mild, moderate and severe based on non-polysomnographic variables: 1) clinical data, 2) symptoms and 3) physical examination.

Methods: In order to produce classification models for OSA severity, five different machine learning methods (Bayesian network, Decision Tree, Random Forest, Neural Networks and Logistic Regression) were trained while relevant variables and their relationships were derived empirically from observed data. Each model was trained and evaluated using 10-fold cross-validation and to evaluate classification performances of all methods, true positive rate (TPR), false positive rate (FPR), Positive Predictive Value (PPV), F measure and Area Under Receiver Operating Characteristics curve (ROC-AUC) were used.

Results: Results of 10-fold cross validated tests with different variable settings promisingly indicated that the OSA severity of suspected OSA patients can be classified, using non-polysomnographic features, with 0.71 true positive rate as the highest and, 0.15 false positive rate as the lowest, respectively. Moreover, the test results of different variables settings revealed that the accuracy of the classification models was significantly improved when physical examination variables were added to the model.

Conclusions: Study results showed that machine learning methods can be used to estimate the probabilities of no, mild, moderate, and severe obstructive sleep apnea and such approaches may improve accurate initial OSA screening and help referring only the suspected moderate or severe OSA patients to sleep laboratories for the expensive tests.

Evaluación del síndrome de apnea-hipopnea obstructiva del sueño (SAHOS) mediante instrumentos de medición como escalas y fórmulas matemáticas

La psicometría del síndrome de apnea-hipopnea obstructiva del sueño (SAHOS) se puede proporcionar con el uso de variados métodos de evaluación como entrevistas clínicas, escalas, cuestionarios de sueño, autoregistros y registros psicofisiológicos. La prueba de oro para el diagnóstico de esta enfermedad sigue siendo la polisomnografía, la cual puede llegar a tener altos costos y dificultades para acceder al estudio. Debido a la alta morbimortalidad asociada a este síndrome, se requieren instrumentos que permitan la identificación rápida de individuos que puedan estar en riesgo de padecerlo. Por tales motivos, se han desarrollado herramientas que permiten detectar los pacientes en riesgo de presentar SAHOS, tales como el Cuestionario de Berlín, el STOP-Bang y la Escala de Somnolencia de Epworth. Es importante tener en cuenta los alcances y limitaciones de estas herramientas para escoger el instrumento indicado según lo que se desee evaluar.

Apnea-hypopnea index prediction through an assessment of autonomic influence on heart rate in wakefulness

With the high prevalence of obstructive sleep apnea, the issue of developing a practical tool for obstructive sleep apnea screening has been raised. Conventional obstructive sleep apnea screening tools are limited in their ability to help clinicians make rational decisions due to their inability to predict the apnea-hypopnea index. Our study aimed to develop a new prediction model that can provide a reliable apnea-hypopnea index value during wakefulness. We hypothesized that patients with more severe obstructive sleep apnea would exhibit more attenuated waking vagal tone, which may result in lower effectiveness in decreasing heart rate as a response to deep inspiration breath-holding. Prior to conducting nocturnal in-laboratory polysomnography, 30 non-obstructive sleep apnea (apnea-hypopnea index<5events/h) subjects and 246 patients with obstructive sleep apnea participated in a 75-second experiment that consisted of a 60-second baseline measurement and consecutive 15-second deep inspiration breath-hold sessions. Two apnea-hypopnea index predictors were devised by considering the vagal activities reflected in the electrocardiographic recordings acquired during the experiment. Using the predictors obtained from 184 individuals, regression analyses and k-fold cross-validation tests were performed to develop an apnea-hypopnea index prediction model. For the remaining 92 individuals, the developed model provided an absolute error (mean±SD) of 3.53±2.67events/h and a Pearson's correlation coefficient of 0.99 (P<0.01) between the apnea-hypopnea index predictive values and the reference values reported by polysomnography. Our study is the first to achieve reliable and time-efficient prediction of the apnea-hypopnea index during wakefulness.

A prediction model based on artificial neural networks for the diagnosis of obstructive sleep apnea

BACKGROUND

Recently, artificial neural networks (ANNs) have been widely applied in science, engineering, and medicine. In the present study, we evaluated the ability of artificial neural networks to be used as a computer program and assistant tool in the diagnosis of obstructive sleep apnea (OSA). Our hypothesis was that ANNs could use clinical information to precisely predict cases of OSA.

METHOD

The study population in this clinical trial consisted of 201 patients with suspected OSA (140 with a positive diagnosis of OSA and 61 with a negative diagnosis of OSA). The artificial neural network was trained by assessing five clinical variables from 201 patients; efficiency was then estimated in this group of 201 patients. The patients were classified using a five-element input vector. ANN classifiers were assessed with the multilayer perceptron (MLP) networks.

RESULTS

Use of the MLP classifiers resulted in a diagnostic accuracy of 86.6 %, which in clinical practice is high enough to reduce the number of patients evaluated by polysomnography (PSG), an expensive and limited diagnostic resource.

CONCLUSIONS

By establishing a pattern that allows the recognition of OSA, ANNs can be used to identify patients requiring PSG.

Cost minimization using an artificial neural network sleep apnea prediction tool for sleep studies

RATIONALE

More than a million polysomnograms (PSGs) are performed annually in the United States to diagnose obstructive sleep apnea (OSA). Third-party payers now advocate a home sleep test (HST), rather than an in-laboratory PSG, as the diagnostic study for OSA regardless of clinical probability, but the economic benefit of this approach is not known.

OBJECTIVES

We determined the diagnostic performance of OSA prediction tools including the newly developed OSUNet, based on an artificial neural network, and performed a cost-minimization analysis when the prediction tools are used to identify patients who should undergo HST.

METHODS

The OSUNet was trained to predict the presence of OSA in a derivation group of patients who underwent an in-laboratory PSG (n = 383). Validation group 1 consisted of in-laboratory PSG patients (n = 149). The network was trained further in 33 patients who underwent HST and then was validated in a separate group of 100 HST patients (validation group 2). Likelihood ratios (LRs) were compared with two previously published prediction tools. The total costs from the use of the three prediction tools and the third-party approach within a clinical algorithm were compared.

MEASUREMENTS AND MAIN RESULTS

The OSUNet had a higher +LR in all groups compared with the STOP-BANG and the modified neck circumference (MNC) prediction tools. The +LRs for STOP-BANG, MNC, and OSUNet in validation group 1 were 1.1 (1.0-1.2), 1.3 (1.1-1.5), and 2.1 (1.4-3.1); and in validation group 2 they were 1.4 (1.1-1.7), 1.7 (1.3-2.2), and 3.4 (1.8-6.1), respectively. With an OSA prevalence less than 52%, the use of all three clinical prediction tools resulted in cost savings compared with the third-party approach.

CONCLUSIONS

The routine requirement of an HST to diagnose OSA regardless of clinical probability is more costly compared with the use of OSA clinical prediction tools that identify patients who should undergo this procedure when OSA is expected to be present in less than half of the population. With OSA prevalence less than 40%, the OSUNet offers the greatest savings, which are substantial when the number of sleep studies done annually is considered.

Screening for obstructive sleep apnoea in obesity and diabetes--potential for future approaches

BACKGROUND

It is recognised that sleep-disordered breathing (SDB), in particular, obstructive sleep apnoea (OSA) is associated with obesity and diabetes. The complications of OSA include dysregulation of metabolic and cardiovascular homeostasis. With the growing population of diabetes and obesity globally, it is becoming apparent that identifying and screening patients who are at risk is becoming increasingly crucial. Many patients may remain unaware of the potential diagnosis and continue to be undiagnosed. The high prevalence of OSA poses a demanding challenge to healthcare providers in order to provide sufficient resources and facilities for patient diagnosis and treatment.

DESIGN

In this article, we review the evidence in favour of screening populations deemed to be at increased risk of OSA, with particular reference to patients with obesity and diabetes. We consider the recent advances in potential screening methods that may allow new prognostic and predictive tools to be developed. A detailed search of Medline and Web of Science electronic databases for relevant articles in English was performed.

RESULTS

Apart from the use of screening tools such as questionnaires and clinical decision models, there is increasing evidence to suggest that there are differences in biological parameters in patients with OSA. Although further studies are required, there may be potential for such biomarkers to contribute to and augment the screening process. However, the significance of such biological tools remains to be elucidated.

CONCLUSIONS

A fundamental role for improved screening in patients with conditions such as obesity and diabetes can enable early interventions that may improve health outcomes relating to the adverse consequences of OSA. The future will see further research being carried out in the development of potential screening methods with emphasis on the selection of patients at risk of sleep disorders, thereby allowing more detailed physiological studies to be carried out where needed.

A hybrid seasonal prediction model for tuberculosis incidence in China

BACKGROUND

Tuberculosis (TB) is a serious public health issue in developing countries. Early prediction of TB epidemic is very important for its control and intervention. We aimed to develop an appropriate model for predicting TB epidemics and analyze its seasonality in China.

METHODS

Data of monthly TB incidence cases from January 2005 to December 2011 were obtained from the Ministry of Health, China. A seasonal autoregressive integrated moving average (SARIMA) model and a hybrid model which combined the SARIMA model and a generalized regression neural network model were used to fit the data from 2005 to 2010. Simulation performance parameters of mean square error (MSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) were used to compare the goodness-of-fit between these two models. Data from 2011 TB incidence data was used to validate the chosen model.

RESULTS

Although both two models could reasonably forecast the incidence of TB, the hybrid model demonstrated better goodness-of-fit than the SARIMA model. For the hybrid model, the MSE, MAE and MAPE were 38969150, 3406.593 and 0.030, respectively. For the SARIMA model, the corresponding figures were 161835310, 8781.971 and 0.076, respectively. The seasonal trend of TB incidence is predicted to have lower monthly incidence in January and February and higher incidence from March to June.

CONCLUSIONS

The hybrid model showed better TB incidence forecasting than the SARIMA model. There is an obvious seasonal trend of TB incidence in China that differed from other countries.

Prediagnosis of obstructive sleep apnea via multiclass MTS

Obstructive sleep apnea (OSA) has become an important public health concern. Polysomnography (PSG) is traditionally considered an established and effective diagnostic tool providing information on the severity of OSA and the degree of sleep fragmentation. However, the numerous steps in the PSG test to diagnose OSA are costly and time consuming. This study aimed to apply the multiclass Mahalanobis-Taguchi system (MMTS) based on anthropometric information and questionnaire data to predict OSA. Implementation results showed that MMTS had an accuracy of 84.38% on the OSA prediction and achieved better performance compared to other approaches such as logistic regression, neural networks, support vector machine, C4.5 decision tree, and rough set. Therefore, MMTS can assist doctors in prediagnosis of OSA before running the PSG test, thereby enabling the more effective use of medical resources.

The STOP-Bang equivalent model and prediction of severity of obstructive sleep apnea: relation to polysomnographic measurements of the apnea/hypopnea index

BACKGROUND

Various models and questionnaires have been developed for screening specific populations for obstructive sleep apnea (OSA) as defined by the apnea/hypopnea index (AHI); however, almost every method is based upon dichotomizing a population, and none function ideally. We evaluated the possibility of using the STOP-Bang model (SBM) to classify severity of OSA into 4 categories ranging from none to severe.

METHODS

Anthropomorphic data and the presence of snoring, tiredness/sleepiness, observed apneas, and hypertension were collected from 1426 patients who underwent diagnostic polysomnography. Questionnaire data for each patient was converted to the STOP-Bang equivalent with an ordinal rating of 0 to 8. Proportional odds logistic regression analysis was conducted to predict severity of sleep apnea based upon the AHI: none (AHI < 5/h), mild (AHI ≥ 5 to < 15/h), moderate (≥ 15 to < 30/h), and severe (AHI ≥ 30/h).

RESULTS

Linear, curvilinear, and weighted models (R(2) = 0.245, 0.251, and 0.269, respectively) were developed that predicted AHI severity. The linear model showed a progressive increase in the probability of severe (4.4% to 81.9%) and progressive decrease in the probability of none (52.5% to 1.1%). The probability of mild or moderate OSA initially increased from 32.9% and 10.3% respectively (SBM score 0) to 39.3% (SBM score 2) and 31.8% (SBM score 4), after which there was a progressive decrease in probabilities as more patients fell into the severe category.

CONCLUSIONS

The STOP-Bang model may be useful to categorize OSA severity, triage patients for diagnostic evaluation or exclude from harm.

Classification algorithms for predicting sleepiness and sleep apnea severity

Identifying predictors of subjective sleepiness and severity of sleep apnea are important yet challenging goals in sleep medicine. Classification algorithms may provide insights, especially when large data sets are available. We analyzed polysomnography and clinical features available from the Sleep Heart Health Study. The Epworth Sleepiness Scale and the apnea-hypopnea index were the targets of three classifiers: k-nearest neighbor, naive Bayes and support vector machine algorithms. Classification was based on up to 26 features including demographics, polysomnogram, and electrocardiogram (spectrogram). Naive Bayes was best for predicting abnormal Epworth class (0-10 versus 11-24), although prediction was weak: polysomnogram features had 16.7% sensitivity and 88.8% specificity; spectrogram features had 5.3% sensitivity and 96.5% specificity. The support vector machine performed similarly to naive Bayes for predicting sleep apnea class (0-5 versus >5): 59.0% sensitivity and 74.5% specificity using clinical features and 43.4% sensitivity and 83.5% specificity using spectrographic features compared with the naive Bayes classifier, which had 57.5% sensitivity and 73.7% specificity (clinical), and 39.0% sensitivity and 82.7% specificity (spectrogram). Mutual information analysis confirmed the minimal dependency of the Epworth score on any feature, while the apnea-hypopnea index showed modest dependency on body mass index, arousal index, oxygenation and spectrogram features. Apnea classification was modestly accurate, using either clinical or spectrogram features, and showed lower sensitivity and higher specificity than common sleep apnea screening tools. Thus, clinical prediction of sleep apnea may be feasible with easily obtained demographic and electrocardiographic analysis, but the utility of the Epworth is questioned by its minimal relation to clinical, electrocardiographic, or polysomnographic features.

A novel approach to prediction of mild obstructive sleep disordered breathing in a population-based sample: the Sleep Heart Health Study

This manuscript considers a data-mining approach for the prediction of mild obstructive sleep disordered breathing, defined as an elevated respiratory disturbance index (RDI), in 5,530 participants in a community-based study, the Sleep Heart Health Study. The prediction algorithm was built using modern ensemble learning algorithms, boosting in specific, which allowed for assessing potential high-dimensional interactions between predictor variables or classifiers. To evaluate the performance of the algorithm, the data were split into training and validation sets for varying thresholds for predicting the probability of a high RDI (≥7 events per hour in the given results). Based on a moderate classification threshold from the boosting algorithm, the estimated post-test odds of a high RDI were 2.20 times higher than the pre-test odds given a positive test, while the corresponding post-test odds were decreased by 52% given a negative test (sensitivity and specificity of 0.66 and 0.70, respectively). In rank order, the following variables had the largest impact on prediction performance: neck circumference, body mass index, age, snoring frequency, waist circumference, and snoring loudness.

Early diagnosis of sleep related breathing disorders

Obstructive sleep apnea (OSA) being the most frequent sleep related breathing disorder results in non-restorative sleep, an increased cardiovascular morbidity and mortality as well as an elevated number of accidents. In Germany at least two million people have to be expected. If obstructive sleep apnea is diagnosed early enough then sleep may regain its restorative function, daytime performance may be improved and accident risk as well as cardiovascular risk may be normalised. This review critically evaluates anamnestic parameters, questionnaires, clinical findings and unattended recordings during sleep regarding their diagnostic accurracy in recognising OSA. There are numerous tools with insufficient results or too few data disqualifying them for screening for OSA. Promising preliminary results are published concerning neural network analysis of a high number of clinical parameters and non-linear analysis of oximetry itself or in combination with heart rate. Nasal pressure recordings can be used for risk estimation even without expertise in sleep medicine. More data is needed. Unattended portable monitoring used by qualified physicians is the gold standard procedure when screening methods for OSA are compared. It has a very high sensitivity and specificity well documented by several meta-analyses.

Obstructive sleep apnea syndrome: natural history, diagnosis, and emerging treatment options

Sleep apnea is an entity characterized by repetitive upper airway obstruction resulting in nocturnal hypoxia and sleep fragmentation. It is estimated that 2%-4% of the middle-aged population has sleep apnea with a predilection in men relative to women. Risk factors of sleep apnea include obesity, gender, age, menopause, familial factors, craniofacial abnormalities, and alcohol. Sleep apnea has been increasingly recognized as a major health burden associated with hypertension and increased risk of cardiovascular disease and death. Increased airway collapsibility and derangement in ventilatory control responses are the major pathological features of this disorder. Polysomnography (PSG) is the gold-standard method for diagnosis of sleep apnea and assessment of sleep apnea severity; however, portable sleep monitoring has a diagnostic role in the setting of high pretest probability sleep apnea in the absence of significant comorbidity. Positive pressure therapy is the mainstay therapy of sleep apnea. Other treatment modalities, such as upper airway surgery or oral appliances, may be used for the treatment of sleep apnea in select cases. In this review, we focus on describing the sleep apnea definition, risk factor profile, underlying pathophysiologic mechanisms, associated adverse consequences, diagnostic modalities, and treatment strategies.

Prediction of obstructive sleep apnea syndrome in a large Greek population

PURPOSE

We aimed to evaluate the predictive value of anthropometric measurements and self-reported symptoms of obstructive sleep apnea syndrome (OSAS) in a large number of not yet diagnosed or treated patients. Commonly used clinical indices were used to derive a prediction formula that could identify patients at low and high risk for OSAS.

METHODS

Two thousand six hundred ninety patients with suspected OSAS were enrolled. We obtained weight; height; neck, waist, and hip circumference; and a measure of subjective sleepiness (Epworth sleepiness scale--ESS) prior to diagnostic polysomnography. Excessive daytime sleepiness severity (EDS) was coded as follows: 0 for ESS ≤ 3 (normal), 1 for ESS score 4-9 (normal to mild sleepiness), 2 for score 10-16 (moderate to severe sleepiness), and 3 for score >16 (severe sleepiness). Multivariate linear and logistic regression analysis was used to identify independent predictors of apnea-hypopnea index (AHI) and derive a prediction formula.

RESULTS

Neck circumference (NC) in centimeters, body mass index (BMI) in kilograms per square meter, sleepiness as a code indicating EDS severity, and gender as a constant were significant predictors for AHI. The derived formula was: AHIpred = NC × 0.84 + EDS × 7.78 + BMI × 0.91 - [8.2 × gender constant (1 or 2) + 37]. The probability that this equation predicts AHI greater than 15 correctly was 78%.

CONCLUSIONS

Gender, BMI, NC, and sleepiness were significant clinical predictors of OSAS in Greek subjects. Such a prediction formula can play a role in prioritizing patients for PSG evaluation, diagnosis, and initiation of treatment.

Derivation and validation of a simple perioperative sleep apnea prediction score

BACKGROUND

Obstructive sleep apnea (OSA) is a largely underdiagnosed, common condition, which is important to diagnose preoperatively because it has implications for perioperative management. Our purpose in this study was to identify independent clinical predictors of a diagnosis of OSA in a general surgical population, develop a perioperative sleep apnea prediction (P-SAP) score based on these variables, and validate the P-SAP score against standard overnight polysomnography.

METHODS

A retrospective, observational study was designed to identify patients with a known diagnosis of OSA. Independent predictors of a diagnosis of OSA were derived by logistic regression, based on which prediction tool (P-SAP score) was developed. The P-SAP score was then validated in patients undergoing overnight polysomnography.

RESULTS

The P-SAP score was derived from 43,576 adult cases undergoing anesthesia. Of these, 3884 patients (7.17%) had a documented diagnosis of OSA. Three demographic variables: age > 43 years, male gender, and obesity; 3 history variables: history of snoring, diabetes mellitus Type 2, and hypertension; and 3 airway measures: thick neck, modified Mallampati class 3 or 4, and reduced thyromental distance were identified as independent predictors of a diagnosis of OSA. A diagnostic threshold P-SAP score > or = 2 showed excellent sensitivity (0.939) but poor specificity (0.323), whereas for a P-SAP score > or = 6, sensitivity was poor (0.239) with excellent specificity (0.911). Validation of this P-SAP score was performed in 512 patients with similar accuracy.

CONCLUSION

The P-SAP score predicts diagnosis of OSA with dependable accuracy across mild to severe disease. The elements of the P-SAP score are derived from a typical university hospital surgical population.

Single layer network classifiers to assist in the detection of obstructive sleep apnea syndrome from oximetry data

The aim of this study is to assess the utility of single layer network classifiers to help in the diagnosis of the obstructive sleep apnea syndrome (SAOS). Oxygen saturation (SaO(2)) recordings from a total of 157 subjects suspected of suffering from OSAS were used. These were divided into a training set and a test set with 74 and 83 subjects, respectively. Four classification schemes were developed by using generalized linear models (GLM). Two GLM classifiers were built with spectral (GLM-SP) and non-linear (GLM-NL) features from SaO(2) signals, respectively. In addition, both algorithms were combined in order to improve their classification capability. The performance of two different ensemble classifiers was analyzed. The highest diagnostic accuracy was reached by the GLM-SP classifier (88%). The ensemble built from the combination of GLM-SP and GLM-NL by means of an additional GLM structure provided the best sensitivity value (87.8%). Applying spectral and non-linear features from SaO(2) data simultaneously could be useful in OSAS diagnosis.

Utility of multilayer perceptron neural network classifiers in the diagnosis of the obstructive sleep apnoea syndrome from nocturnal oximetry

The aim of this study is to assess the ability of multilayer perceptron (MLP) neural networks as an assistant tool in the diagnosis of the obstructive sleep apnoea syndrome (OSAS). Non-linear features from nocturnal oxygen saturation (SaO(2)) recordings were used to discriminate between OSAS positive and negative patients. A total of 187 subjects suspected of suffering from OSAS (111 with a positive diagnosis of OSAS and 76 with a negative diagnosis of OSAS) took part in the study. The initial population was divided into training, validation and test sets for deriving and testing our neural network classifier. Three methods were applied to extract non-linear features from SaO(2) signals: approximate entropy (ApEn), central tendency measure (CTM) and Lempel-Ziv complexity (LZC). The selected MLP-based classifier provided a diagnostic accuracy of 85.5% (89.8% sensitivity and 79.4% specificity). Our neural network algorithm could represent a useful technique for OSAS detection. It could contribute to reduce the demand for polysomnographic studies in OSAS screening.

Applying neural network classifiers in the diagnosis of the obstructive sleep apnea syndrome from nocturnal pulse oximetric recordings

The aim of this study was to assess the ability of neural networks as an assistant tool for the diagnosis of the obstructive sleep apnea syndrome (OSAS). A total of 187 subjects suspected of suffering from OSAS (111 with a positive diagnosis of OSAS and 76 with a negative diagnosis of OSAS) took part in the study. The initial population was divided into training, validation and test sets for deriving and testing our neural classifiers. Our method was based on spectral and non-linear features extracted from overnight arterial oxygen saturation (SaO_(2)) recordings. A seven-element input vector was used for patient classification. We selected four spectral features from the estimated power spectral density (PSD) of SaO_(2). In addition, three input features were computed from non-linear analysis of SaO_(2). Two neural classifiers were assessed: the multilayer perceptron (MLP) network and the radial basis function (RBF) network. The RBF classifier provided the best diagnostic performance with an accuracy of 86.3% (89.9% sensitivity and 81.1% specificity).

Radial basis function classifiers to help in the diagnosis of the obstructive sleep apnoea syndrome from nocturnal oximetry

The aim of this study is to assess the ability of radial basis function (RBF) classifiers as an assistant tool for the diagnosis of the obstructive sleep apnoea syndrome (OSAS). A total of 187 subjects suspected of suffering from OSAS were available for our research. The initial population was divided into training, validation and test sets for deriving and testing our neural classifiers. We used nonlinear features from nocturnal oxygen saturation (SaO(2)) to perform patients' classification. We evaluated three different RBF construction techniques based on the following algorithms: k-means (KM), fuzzy c-means (FCM) and orthogonal least squares (OLS). A diagnostic accuracy of 86.1, 84.7 and 85.5% was provided by the networks developed with KM, FCM and OLS, respectively. The three proposed networks achieved an area under the receiver operating characteristic (ROC) curve over 0.90. Our results showed that a useful non-invasive method could be applied to diagnose OSAS from nonlinear features of SaO(2) with RBF classifiers.

Sleep apnea detection and classification using fuzzy logic: clinical evaluation

We have previously reported a system suitable for detection and classification of sleep apnea syndromes. This paper reports the results of the clinical evaluation of the proposed system. In the current implementation, the system uses breathing signals: nasal flow, thorax movement, and abdomen movement. The detection part of the system uses only the nasal flow signal to detect apnea employing two engines used in series. It then feeds segments labeled as abnormal to the classification part of the system, which uses the center of gravity of each segment to determine the type of abnormality: obstructive, central or hypopnea. In comparison to other systems, this implementation can be shown to be simpler and more accurate. When the low implementation cost is taken into consideration, the proposed system has a substantial potential for being used as a screening device.

Sleep apnea detection using an adaptive fuzzy logic based screening system

We report an adaptive diagnostic system for the classification of breathing events for the purpose of detecting sleep apnea syndromes. The system employs two classification engines used in series. The first engine is fuzzy logic-based and generates one of three outcomes for each breathing event: normal, abnormal, and not-sure. The second classification engine is based on a center of gravity engine which is trained using the normal and abnormal events, generated by the first engine, and is specifically designed for sorting out the not-sure events. The fuzzy logic engine can be tuned very conservatively to reduce or eliminate the chance of error at the first stage. Since the second engine is trained adaptively using normal and abnormal data of the same patient, its accuracy is generally better than relying on multi-patient training approaches. The two-step, adaptive nature of the system allows for high accuracy and lends itself well for practical implementation.

The predictive value of clinical and epidemiological parameters in the identification of patients with obstructive sleep apnoea (OSA): a clinical prediction algorithm in the evaluation of OSA

We sought to analyze the predictive value of anthropometric, clinical and epidemiological parameters in the identification of patients with suspected OSA, and their relationship with apnoea/hypopnoea respiratory events during sleep. We studied retrospectively 433 patients with OSA, 361 men (83.37%) and 72 women (16.63%), with an average age of +/-47, standard deviation +/-11.10 years (range 18-75 years). The study variables for all of the patients were age, sex, spirometry, neck circumference, body mass index (BMI), Epworth sleepiness scale, nasal examination, pharyngeal examination, collapsibility of the pharynx (Müller Manoeuvre), and apnoea-hypopnoea index (AHI). Age, neck circumference, BMI, Epworth sleepiness scale, pharyngeal examination and pharyngeal collapse were the significant variables. Of the patients, 78% were correctly classified, with a sensitivity of 74.6% and a specificity of 66.3%. We found a direct relationship between the variables analysed and AHI. Based on these results, we obtained the following algorithm to calculate the prediction of AHI for a new patient: AHI = -12.04 + 0.36 neck circumference +2.2286 pharyngeal collapses (MM) + 0.1761 Epworth + 0.0017 BMI x age + 1.1949 pharyngeal examinations. The ratio variance in the number of respiratory events explained by the model was 33% (r2 = 0.33). The variables given in the algorithm are the best ones for predicting the number of respiratory events during sleep in patients studied for suspected OSA. The algorithm proposed may be a good screening method to the identification of patients with OSA.

Neural Networks for Ischemic Stroke

BACKGROUND

To have uniform criteria for evaluating populations for prevalence of transient ischemic attack (TIA)/stroke, validated instruments are necessary for objective assessment and classification.

METHODS

Patient responses compatible with symptoms of TIA or ischemic stroke, obtained from participants in a substudy of the Asymptomatic Carotid Atherosclerosis Study, were used to program a neural network for each symptom. Models were designed for rapid classification into 1 of 7 outputs: no event, TIA, or stroke (in left carotid, right carotid, or vertebrobasilar). The networks were then tested by comparing decisions with a validated questionnaire used to access an independent data set of 381 patients.

RESULTS

There were 144 patients who reported sudden speech change, 89 with sudden vision loss, 67 with double vision, 189 with sudden numbness, 223 with episodic dizziness, and 108 with paralysis, for a total of 820 reported symptoms among the 381 patients tested. For each category, an equal number of individuals reporting "No" to these phenomena were randomly selected and analyzed. Neural network classification correlated with the diagnoses made by specially trained stroke clinicians (e.g., all who responded "No" were correctly classified as having no neurologic event). Ten symptomatic patients were misclassified, with the most common reason being incomplete data. After adjustment of the network logic, these misclassifications did not recur.

CONCLUSION

Computer networks can be trained to produce a rapid and accurate classification of TIA or stroke by vascular distribution, enabling screening of populations for assessment of their incidence and prevalence.

Screening for obstructive sleep apnea: an evidence-based analysis

Sleep disordered breathing is a spectrum of diseases that includes snoring, upper airway resistance syndrome, and obstructive sleep apnea (OSA). Obstructive sleep apnea is a common sleep disorder and is estimated to have an incidence of 24% in men and 9% in women. However, many authors believe that up to 93% of women and 82% of men with moderate to severe OSA remain undiagnosed. There is a strong link between sleep disordered breathing and hypertension, believed to be due to sleep fragmentation, intermittent hypoxemia, and increased sympathetic tone, which results in a higher mortality and morbidity rate among these patients. It is therefore desirable to attempt to diagnose all patients with OSA, to institute early treatment intervention, and to prevent development of cardiovascular complications. The gold standard for diagnosing OSA remains the attended overnight level I polysomnogram. However, in view of the limited resources, including limited number of recording beds, high cost, long waiting lists, and labor requirements, many authors have explored the use of clinical predictors or questionnaires that may help to identify higher-risk patients. Screening devices in the form of single or multiple channel monitoring systems have also been introduced and may represent an alternative method to diagnose OSA. The ideal screening device should be cheap, readily accessible, easily used with minimal instructions, have no risk or side effects to the patient, and be safe and accurate. We review a variety of clinical predictive formulae and several screening devices available for the diagnosis of OSA.

Tracking the states of a nonlinear and nonstationary system in the weight-space of artificial neural networks

We propose a novel interpretation and usage of Neural Network (NN) in modeling physiological signals, which are allowed to be nonlinear and/or nonstationary. The method consists of training a NN for the k-step prediction of a physiological signal, and then examining the connection-weight-space (CWS) of the NN to extract information about the signal generator mechanism. We define a novel feature, Normalized Vector Separation (gamma(ij)), to measure the separation of two arbitrary states "i" and "j" in the CWS and use it to track the state changes of the generating system. The performance of the method is examined via synthetic signals and clinical EEG. Synthetic data indicates that gamma(ij) can track the system down to a SNR of 3.5 dB. Clinical data obtained from three patients undergoing carotid endarterectomy of the brain showed that EEG could be modeled (within a root-means-squared-error of 0.01) by the proposed method, and the blood perfusion state of the brain could be monitored via gamma(ij), with small NNs having no more than 21 connection weight altogether.

The association of tongue scalloping with obstructive sleep apnea and related sleep pathology

OBJECTIVE

The association between OSAS and patient history and physical exam findings is previously established; however, to our knowledge there are no studies that evaluate the role of tongue scalloping as a reliable clinical indicator for OSA, snoring, or the presence of other sleep pathology as evidenced by polysomnography. This study evaluates the hypothesis that such an association exists.

SUBJECTS AND METHODS

Sixty-one otolaryngology clinic patients were evaluated by history and physical exam for the presence and degree of tongue scalloping, snoring, and other previously established clinical indicators for sleep-disordered breathing and obstructive apnea. Twenty-five of the 61 study patients were additionally evaluated by overnight polysomnography to provide conclusive diagnosis of sleep pathology. The degree of tongue scalloping was graded from 0 to 3 and its significance as a screening, diagnostic, and predictive factor for sleep pathology was then statistically determined.

RESULTS

Twenty-seven patients (44%) had known or newly documented OSA and 47 (77%) had a history of snoring. Twenty-seven patients (44%) had some degree of tongue scalloping (1-3) and 74% of these patients were male. The presence of any degree of tongue scalloping (grade 1-3) in patients with known or newly documented OSA showed sensitivity, specificity, PPV, and NPV of 52%, 68%, 70%, and 50% respectively. The presence of tongue scalloping in patients with either known snoring history or newly documented snoring showed sensitivity, specificity, PPV, and NPV of 47%, 64%, 81%, and 26% respectively. Presence of tongue scalloping was 71% specific for abnormal sleep efficiency (<85%), 70% specific for abnormal AHI (>5), and 86% specific for nocturnal desaturation >4% below baseline. Presence of tongue scalloping also showed PPV of 67% for abnormal AHI, 89% for apnea or hypopnea, and 89% for nocturnal desaturation. Presence and severity of tongue scalloping showed positive correlation with increasing Mallampati and modified Mallampati airway classification.

CONCLUSIONS

In high-risk patients we found tongue scalloping to be predictive of sleep pathology. Tongue scalloping was also associated with pathologic polysomnography data and abnormal Mallampati grades. We feel the finding of tongue scalloping is a useful clinical indicator of sleep pathology and that its presence should prompt the physician to inquire about snoring history.

A computer-based approach to assess the perception of composite odour intensity: a step towards automated olfactometry calibration

The 2004 Nobel Prize in Physiology or Medicine laureates, Richard Axel and Linda Buck, have made smell a less enigmatic sense to study. In clinical routine, olfactory function is assessed using defined concentrations of a single defined substance, a setting which is uncommon in daily life. The present study was therefore conducted to evaluate the applicability of composite odours. Air was contaminated with different quantities of cyclohexanol, cyclohexanone and cyclohexane to generate 73 gas mixtures (one component: n = 21, two components: n = 40, three components: n = 12). The intensity of perception was estimated for each mixture by an average of 60.3 healthy individuals (4,403 assessments). An artificial neural network (ANN) was trained and validated using the contaminants' concentrations with the corresponding estimated intensities. The inter-rater variability was low, as 75.7% of the assessments did not exceed a difference beyond 0.5 from the corresponding median (considered correct predictions). The ANN correctly estimated 78.1% of the gas mixtures, and in terms of the regression task the ANN demonstrated a sufficient prediction performance (Pearson's correlation coefficient r = 0.883; R(2) = 0.757) and outperformed linear regression (r = 0.770; R(2) = 0.667). Evaluating extra ANNs for gas mixtures comprising one, two or three components, the predictive power did not decrease when complexity increased. The aforementioned results reflect nonlinearity in human perception. ANN technology helps simulate human perception of composite odour intensity which may be applicable to olfactometry calibration and systems biological mathematical modelling. The use of composite odours may represent real-life problems more adequately than single substances.

Prediction of obstructive sleep apnea by OxiFlow in overweight patients

OBJECTIVES

The aim of our study was to assess the diagnostic characteristics of the OxiFlow (OF) device that combines oximetry with recording of thermistor airflow.

METHODS

In patients referred to the sleep laboratory of an obesity clinic apnea-hypopnea index (AHI, events h(-1)) was calculated both by a full-night polysomnography (PSG) and OF on a separate night. Fifty-six patients were studied, of whom 49 had OSA defined as an AHI> or =15 events h(-1).

RESULTS

There was an underestimation of AHI by OF when assessed by the Bland-Altman plot. Sensitivity (Se), specificity (Sp), positive (PPV) and negative (NPV) predictive values for OF-AHI thresholds (10, 15 and 20 events h(-1)), taking PSG as a gold standard with a fixed PSG-AHI threshold of 15 events h(-1), were evaluated in two groups of patients with intermediate (group A, n=18, OSA prevalence=72.2%) and high (group B, n=38, OSA prevalence=94.7%) clinical probability of OSA. Se and PPV ranged respectively from 0.77 to 0.85 and from 0.73 to 0.77 (group A); from 0.74 to 0.97 and from 0.94 to 0.98 (group B). Sp and NPV ranged respectively from 0.20 to 0.40 and from 0.33 to 0.40 (group A); from 0.50 to 0.83 and from 0.21 to 0.67 (group B). Likelihood ratios (LRs) for a positive OF result ranged from 1.06 to 1.28 (group A) and from 1.83 to 4.42 (group B).

CONCLUSIONS

We conclude that in a population with a high OSA prevalence, we have found a low agreement between PSG-AHI and OF-AHI and an underestimation of AHI by OF. The LRs of OF as a diagnostic test were of low significance, precluding its usefulness in generating significant shifts in pretest to posttest probability of OSA.