<p>Cardiovascular disease (CVD) remains the leading cause of global mortality, yet early detection is often limited by episodic, resource-intensive clinical assessments that are poorly suited for population-level screening. Sleep offers a uniquely informative physiological window for cardiovascular monitoring. Recent studies have demonstrated the promise of deep-learning models applied to sleep data for CVD screening. However, most existing approaches rely predominantly on electrocardiography (ECG), which is not routinely available in scalable home sleep monitoring (HSM), thereby constraining their practical deployment. This study proposes HSM-CVD-Net, a multimodal ECG-free deep-learning framework for segment-level CVD detection using standard HSM signals. The model integrates abdominal respiratory effort, peripheral oxygen saturation, and pulse rate to capture nocturnal cardiorespiratory interactions associated with cardiovascular dysfunction. Developed and evaluated using data from the Sleep Heart Health Study, the framework was formulated as a binary classification task for specific CVD subtypes. HSM-CVD-Net achieved segment-level accuracies of 88.61%, 90.81%, 91.94%, and 93.93%, with corresponding F1-scores of 88.84%, 90.77%, 92.06%, and 94.06% for myocardial infarction, congestive heart failure, angina, and stroke detection, respectively. These findings indicate that high-performance classification of CVD subtypes can be achieved using a minimal, ECG-free sensor configuration, extending the clinical utility of sleep monitoring beyond sleep apnea and enabling scalable population-level cardiovascular screening.</p>