Sleep apnea syndrome (SAS) is a nocturnal respiratory disorder that can be associated with long-term cardiovascular complications. Alternative screening solutions are currently being developed to overcome the limitations of reference in-lab polysomnography. As the respiratory signal can be reconstructed from the electrocardiogram (ECG), the latter is all the more interesting as its recording is easy and non-invasive for the patient. The application of deep learning algorithms using ECGs has proved effective in classifying sleep-related pathological events. In this paper, we propose a novel hybrid architecture to detect apneic episodes using single-lead ECGs. Following a preprocessing step, morphological and temporal components of interest are extracted through convolutional and recurrent blocks, respectively. Additional mechanisms are further integrated to enhance the classification. Models were trained and validated on a dataset derived from STAGES and Apnea-ECG databases. Influence of patient phenotype on classification was estimated by comparing the performance between several groups of patients with different clinical information. Overall, a model we have developed performs competitively with the best current methods by accurately classifying patients to different degrees of severity with average sensitivity, specificity and accuracy of 93.38%, 75.46% and 86.66%, respectively.