A simple, efficient algorithm has been developed that will help classify symptom subtypes among patients with obstructive sleep apnea (OSA; minimally symptomatic, disturbed sleep, moderately sleepy, and excessively sleepy) based on only 6 symptom items. Results on the development and validation of this tool were presented at the 2019 Annual Meeting of the World Sleep Society, which took place from September 20 to 25, 2019, in Vancouver, Canada.

Investigators sought to identify the minimum number of questions needed for the accurate prediction and classification of new patients with OSA. They used latent class analysis of symptom questionnaire data (a total of 13 questions, plus the Epworth Sleepiness Scale [ESS] score) from a total of 1730 patients with moderate to severe OSA (apnea-hypopnea index ≥15), which were available from the Sleep Apnea Global Interdisciplinary Consortium (SAGIC). They trained and evaluated a supervised machine learning method to establish the minimum number of questions needed for accurate prediction of the OSA symptom subtypes.

Each of the 14 symptom items was ranked according to its relevance in predicting the subtypes, with the predictive performance of sequential models assessed by including between 1 and 14 symptom questions, based on order of importance. The optimal number of questions needed to provide the highest acceptable accuracy was determined according to a change in the average balanced accuracy of <1% after inclusion of the next question in the importance ranking. The optimal model was validated in an independent clinical cohort from Iceland (n=785) and a community-based sample from the United States (n=1207).

Results of the analysis showed that the model with the highest acceptable accuracy (average balanced accuracy, 87.8%; average multiclass area under the receiver operating characteristic curve [mAUC], 97.6%) in the testing samples included the following 6 items: ESS, “feel rested during the day,” “difficulty maintaining sleep,” “physically tired,” “sleep involuntarily,” and “sleepy during the day.” The model accurately predicted 94.3% of the patients from SAGIC into the appropriate OSA subtype. When the researchers applied this model in an independent clinical cohort (Iceland) and a community-based sample (United States), the predictive performance was clinically acceptable (mAUC, 90.2%; and mAUC, 84.0%, respectively).

Related Articles

The investigators concluded that the use of this newly developed algorithm will facilitate clinical translation by supporting the efficient identification of patients with the excessively sleepy OSA subtype who are at increased risk for cardiovascular disease.

Reference

Mazzotti D, Keenan B, Kim J, et al. A simple algorithm for accurate identification of obstructive sleep apnea symptom subtypes: validation across international sleep centers and a community-based cohort. Presented at: World Sleep Society 2019 Annual Meeting; September 20-25, 2019; Vancouver, Canada. Abstract 292.