Risk prediction tools are crucial for population-based management of cardiovascular disease (CVD). However, most prediction models are currently used to assess long-term risk instead of the risk of short-term CVD onset. We developed a Dynamic Risk-based Early wArning Monitoring (DREAM) system using large-scale, real-time electronic health record data from 2010 to 2020 from the CHinese Electronic health Records Research in Yinzhou study. The dynamic risk scores were derived from a 1:5 matched nested case–control set comprising 70,470 individuals (11,745 CVD events) and then validated in a cohort of 81,205 individuals (5950 CVD events). The individuals were Chinese adults aged 40–79 years without a history of CVD at baseline. Eleven predictors related to vital signs, laboratory tests, and health service utilization were selected to establish the dynamic scores. The proposed scores were significantly associated with the subsequent CVD onset (adjusted odds ratio, 1.21; 95% confidence interval, 1.20–1.23). The area under the receiver operating characteristic curves (AUCs) was 0.6010 (0.5929–0.6092) and 0.6021 (0.5937–0.6105) for the long-term 10-year CVD risk <10% and ≥10% groups in the derivation set, respectively. In the long-term 10-year CVD risk ≥10% group in the validation set, the change in AUC in addition to the long-term risk was 0.0235 (0.0155–0.0315). By increasing the risk threshold from 7 to 16 points, the proportion of true subsequent CVD cases among those given alerts increased from 40.61% to 85.31%. In terms of management efficiency, the number needed to manage per CVD case ranged from 2.46 to 1.17 using the risk scores. With the increasing popularity and integration of EHR systems with wearable technology, the DREAM scores can be incorporated into an early-warning system and applied in dynamic, real-time, EHR-based, automated management to support healthcare decision making for individuals, general practitioners, and policymakers.