Chest
Volume 116, Issue 2, August 1999, Pages 409-415
Journal home page for Chest

Clinical Investigations
Sleep and Breathing
Neural Network Prediction of Obstructive Sleep Apnea From Clinical Criteria

https://doi.org/10.1378/chest.116.2.409Get rights and content

Study objectives

Clinical prediction models for the diagnosis of obstructive sleep apnea (OSA) have lacked the accuracy necessary to confidently replace polysomnography (PSG). Artificial neural networks are computer programs that can be trained to predict outcomes based on experience. This study was conducted to test the hypothesis that a generalized regression neural network (GRNN) could accurately classify patients with OSA from clinical data.

Study design

Retrospective review.

Setting

Regional sleep referral center.

Patients

Randomly selected records of patients referred for possible OSA.

Measurements

The neural network was trained using 23 clinical variables from 255 patients, and the predictive performance was evaluated using 150 other patients.

Results

The prevalence of OSA in this series of 405 patients (293 men and 112 women) was 69%. The trained GRNN had an accuracy of 91.3% (95% confidence interval [CI], 86.8 to 95.8). The sensitivity was 98.9% for having OSA (95% CI, 96.7 to 100), and the specificity was 80% (95% CI, 70 to 90). The positive predictive value that the patient would have OSA was 88.1% (95% CI, 81.8 to 94.4), whereas the negative predictive value that the patient would not have OSA (if so classified) was 98% (95% CI, 94 to 100).

Conclusions

Appropriately trained GRNN has the ability to accurately rule in OSA from clinical data, and GRNN did not misclassify patients with moderate to severe OSA. In this study, use of the neural network could have reduced the number of PSG studies performed. Prospective validation of the neural network for the diagnosis of OSA is now required.

Section snippets

Materials and Methods

The clinical data were collected from a retrospective review of randomly selected patients who presented to the London Health Sciences Center Sleep Clinic for assessment of possible OSA and who went on to have PSG. Greater than 95% of patients referred for possible OSA have PSG performed in their work-up. Reasons for not going on to PSG included a clinical diagnosis other than OSA (eg, insomnia or sleep-related laryngospasm) and patient refusal. Excluded from the study were patients < 16 years

Results

The prevalence of OSA (defined as AHI ≥ 10) in this series was 69% (53% in women and 76% in men). The demographic and clinical features of the study population are outlined in Table 1. There were more men than women in the data set, and on average the patients were middle aged and overweight. The prevalence of OSA in the training set and the test set were 72% and 65%, respectively (p > 0.05). There were small differences between the training set and the test set. The subjects in the test set

Discussion

We have shown in this retrospective study of a population of OSA patients that a trained GRNN can accurately diagnose the presence of OSA (defined as AHI ≥ 10/h). To our knowledge, this study is the first use of a neural network to predict the presence or absence of OSA. The prevalence of OSA in this population was 69%, with OSA more common in men than in women. In the overall group, the presence of OSA was associated with a history of witnessed apneas, observed choking, and increased smoking

Conclusion

This study has shown that an optimally trained GRNN can accurately rule out OSA in a referral setting where the prevalence of OSA is high (ie, 69%). Also, the neural network did not miss important cases of OSA. This neural network was developed from readily available clinical data found in most patient charts. Approximately 32% of all PSG studies could have been avoided by using this classification tool. Training with increased numbers of subjects can be anticipated to only improve network

References (26)

  • KA Ferguson et al.

    Sleep-related breathing disorders: 4. Consequences of sleep disordered breathing

    Thorax

    (1995)
  • B Crocker et al.

    Estimation of the probability of disturbed breathing during sleep before a sleep study

    Am Rev Respir Dis

    (1990)
  • S Viner et al.

    Are history and physical examination a good screening test for sleep apnea?

    Ann Intern Med

    (1991)
  • Cited by (79)

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

      2022, Patterns
      Citation Excerpt :

      Machine learning models, in particular neural networks, extract valuable insights from data and have achieved unprecedented predictive performance in the healthcare domain, e.g., in single-cell analysis,5 aiding medical diagnosis and treatment,6,7 or in personalized medicine.8

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

      2021, Sleep Medicine
      Citation Excerpt :

      Recent advances in artificial intelligence (AI) models have led to the development of automated medical diagnostic systems that can aid in the diagnosis of disease states. Various neural networks (NNs) have been proposed and shown promising results in medical diagnostics [18–21], including the prediction of apnea and hypopnea episodes and the diagnosis and severity of OSAHS [22–25]. However, NNs suffer from the difficulty in explaining the inference process behind their black-box-like models.

    • OSAS assessment with entropy analysis of high resolution snoring audio signals

      2020, Biomedical Signal Processing and Control
      Citation Excerpt :

      OSAS prediction is a long-standing challenge. Kirby et al. [39] achieved good results, using a neural network and a single hard threshold (AHI ≥ 10), with a sensitivity of 98.9% and specificity of 80%. Rowley et al. [40] tried to predict OSAS for 2 different AHI thresholds, ≥10 and ≥20.

    • A different sleep apnea classification system with neural network based on the acceleration signals

      2020, Applied Acoustics
      Citation Excerpt :

      Especially Artificial intelligence (AI) methods have been studied in many numbers of papers. Artificial Neural Networks (ANNs) method is one of the AI methods and is widely used in the apnea classification [11,12]. Besides, at the related works in the literature [13,14], Particle Swarm Optimization (PSO) and Machine Learning (ML) techniques are used to find a solution to diagnose apnea.

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

      2017, Physiology and Behavior
      Citation Excerpt :

      Prior to conducting overnight in-laboratory polysomnography (PSG) or cardiorespiratory polygraphy, formal diagnostic tests for OSA, employing screening tools is useful for cost- and time-effective identification of patients with OSA. Most widely used screening tools for OSA are based on clinical predictors, including demographic (e.g., age and gender) and anthropometric (e.g., body mass index (BMI) and neck circumference) variables, observed apnea and hypopnea, and comorbid conditions (e.g., snoring, hypertension, and daytime symptoms, such as tiredness, fatigue, and excessive daytime sleepiness) [1,6–23]. Information regarding these variables is generally investigated through questionnaires and/or interviews.

    View all citing articles on Scopus

    Presented in part at the Meeting of the Eastern Section of the American Laryngological, Rhinological and Otological Society Inc., New York, NY, January, 1998.

    View full text