Machine learning-based detection of sleep-disordered breathing in hypertrophic cardiomyopathy

Hypertrophic cardiomyopathy (HCM) is often concomitant with sleep-disordered breathing (SDB), which can cause adverse cardiovascular events. Although an appropriate approach to SDB prevents cardiac remodelling, detection of concomitant SDB in patients with HCM remains suboptimal. Thus, we aimed to develop a machine learning-based discriminant model for SDB in HCM.

In the present multicentre study, we consecutively registered patients with HCM and performed nocturnal oximetry. The outcome was a high Oxygen Desaturation Index (ODI), defined as 3% ODI >10, which significantly correlated with the presence of moderate or severe SDB. We randomly divided the whole participants into a training set (80%) and a test set (20%). With data from the training set, we developed a random forest discriminant model for high ODI based on clinical parameters. We tested the ability of the discriminant model on the test set and compared it with a previous logistic regression model for distinguishing SDB in patients with HCM.

Among 369 patients with HCM, 228 (61.8%) had high ODI. In the test set, the area under the receiver operating characteristic curve of the discriminant model was 0.86 (95% CI 0.77 to 0.94). The sensitivity was 0.91 (95% CI 0.79 to 0.98) and specificity was 0.68 (95% CI 0.48 to 0.84). When the test set was divided into low-probability and high-probability groups, the high-probability group had a higher prevalence of high ODI than the low-probability group (82.4% vs 17.4%, OR 20.9 (95% CI 5.3 to 105.8), Fisher's exact test p<0.001). The discriminant model significantly outperformed the previous logistic regression model (DeLong test p=0.03).

Our study serves as the first to develop a machine learning-based discriminant model for the concomitance of SDB in patients with HCM. The discriminant model may facilitate cost-effective screening tests and treatments for SDB in the population with HCM.