Diagnosis and prediction of patients with severe obstructive apneas using support vector machine

Obstructive sleep apnea (OSA) is a significant cause of motor vehicle crashes and other chronic diseases. Polysomnography (PSG) is a diagnostic test that a number of physiologic variables are measured and recorded during sleep. Although PSG is treated as the golden standard for diagnosing OSA, it is time-consuming and expensive. Therefore, clinical prediction of high-risk OSA patients using questionnaires and application of cheap home diagnostic devices has a great potential in reducing healthcare cost and in eliminating environmental variation for some patients tested in the sleeping center. A total of 699 patients with possible OSA had been recruited and tested using PSG for overnight attending at the Sleep Center of China Medical University Hospital from Jan. 2004 to Dec. 2005. Subjects with age less than 20 or more than 85 years old were excluded. Therefore only the data obtained from 566 patients were used for further analysis. After statistical analyses and feature selection, support vector machine (SVM) was used to discriminate normal subjects and patients with different stages of severity. The results show that oxygen desaturation index (ODI) alone has the best prediction outcome for the patients with severe OSA with a sensitivity as high as 83.51%. The sensitivity is only 42.86% in discriminating the normal from abnormal subjects. From cost-benefit analysis, we propose that home-styled oximeter can be used for sifting severe patients among all suspected patients and then PSG applied for discriminating normal from mild and moderate subjects. A cost reduction of NT$1563 in average can be achieved for each subject under the current Taiwanese insurance setting.