As the demand for personalization and accurate prediction in medical and health management continues to grow, how to timely identify patients’ health risks and provide corresponding intervention measures has become an important challenge facing the current field of cardiac nursing. To address this issue, this study cites a health risk prediction and intervention optimization system based on a deep neural network (DNN), aiming to improve the accuracy of health risk assessment and the effectiveness of intervention. First, this study collects multi-dimensional patient health data, including physical sign information, historical medical records, laboratory test results, etc., to construct a comprehensive health data set. Then, the DNN algorithm is used to accurately predict the patient's health risks. Through its powerful feature learning ability, DNN can mine potential patterns in complex nonlinear relationships and accurately identify high-risk patients. Finally, combined with the intelligent optimization decision-making model, the system generates personalized intervention plans, including drug adjustments, lifestyle improvements, etc., to further optimize the intervention effect. In the experimental conclusion, after the intervention, the prediction accuracy of the DNN model increased to 89.5%, the recall rate increased to 87.8%, the F1 score reached 88.6%, and the risk reduction rate of high-risk patients reached 15.2%. In addition, the system maintained good real-time performance and responsiveness under different data loads, and the response time was within an acceptable range (for example, the total response time for processing 10,000 patient data was 3.3 s). Through experimental verification, the proposed system has significantly improved the accuracy of health risk prediction and intervention effect compared with traditional methods, and has good real-time processing capabilities, showing strong application prospects.