In search of precision for diagnosis of Obstructive Sleep Apnea

Abstract

Background: Obstructive sleep apnea (OSA) is the most frequent sleep-disorder breathing in the world. The apnea–hypopnea index (AHI) is used for diagnosis and severity of OSA and to make decisions about treatment. However, OSA is a heterogeneous disorder due to its pathogenesis and several clinical expressions.

Objective: To compare the precision of comorbidities, oxygen saturation and arousals as predictors of presence and severity of OSA.

Methods: Electronic Health records of 7,924 patients from high altitude (2600m above sea level) with polysomnography were selected to perform the analysis through logistic regression, random forest, and neural network. The variable “comorbidity” included cardiovascular disease, Diabetes, Obesity, asthma and Chronic obstructive pulmonary disease. The methodology used for data analysis was CRISP-DM including the following phases: problem understanding, data understanding, data preparation, modeling, evaluation, and deployment. A multiclass confusion matrix was made (none, mild, moderate, and severe). Precision, recall and f1-scores were calculated for each test.

Results: Oxygen saturation showed the best f1-score to distinguish between absence and severe OSA for the three tests. The values obtained for f1-score in logistic regression were 0.60 and 0.53 (absence and severe OSA); in Random forest were 0.60 and 0.54 (absence and severe OSA), and in neuronal network were 0.60 and 0.53 (absence and severe OSA. (Figure 1).

Conclusion: Precision medicine in OSA has identified the necessity to research new strategies for accuracy diagnosis and treatment in OSA. The results of this investigation suggest that oxygen saturation could be a good tool for diagnosis of OSA.