A deep-learning model (CXR CVD-Risk) was developed to estimate 10-year risk for major adverse cardiovascular events (MACE) from chest radiographs (CXR). The model was tested against the traditional ASCVD risk score to assess its performance in identifying individuals at high risk for MACE, particularly those whose ASCVD risk score could not be calculated due to missing data. The model was developed using data from the PLCO Cancer Screening Trial and externally validated in a separate cohort of 8869 outpatients with unknown ASCVD risk and 2132 outpatients with known ASCVD risk. In the group with unknown ASCVD risk, those predicted to have a 7.5% or higher 10-year MACE risk by CXR CVD-Risk had a higher adjusted hazard ratio (1.73) for MACE compared to those with lower risk. In the group with known ASCVD risk, CXR CVD-Risk predicted MACE beyond the traditional ASCVD risk score (adjusted HR, 1.88). The model showed improved discrimination and calibration compared to the traditional ASCVD risk score. The CXR CVD-Risk model identified a higher proportion of patients as statin-eligible compared to the traditional ASCVD risk score, particularly in those with unknown ASCVD risk. For example, 47.9% of patients with unknown ASCVD risk were identified as statin-eligible by CXR CVD-Risk, with 11.8% developing MACE, compared to 6.2% of those classified as ineligible. In the group with known ASCVD risk, 37.0% were identified as statin-eligible by CXR CVD-Risk, with 7.6% developing MACE, compared to 3.1% of those classified as ineligible. The study suggests that CXR CVD-Risk could help identify individuals at high risk for MACE who may benefit from primary prevention with statins, even when traditional risk factors are not available. The model could be implemented as an automatic tool in electronic medical records to opportunistically analyze existing CXRs, potentially improving statin eligibility decisions. However, the study has limitations, including its retrospective design and potential overestimation of event rates. The model may not be generalizable to racially and ethnically diverse populations. Despite these limitations, the results indicate that deep learning can provide valuable prognostic information from CXRs, complementing traditional risk assessment methods.A deep-learning model (CXR CVD-Risk) was developed to estimate 10-year risk for major adverse cardiovascular events (MACE) from chest radiographs (CXR). The model was tested against the traditional ASCVD risk score to assess its performance in identifying individuals at high risk for MACE, particularly those whose ASCVD risk score could not be calculated due to missing data. The model was developed using data from the PLCO Cancer Screening Trial and externally validated in a separate cohort of 8869 outpatients with unknown ASCVD risk and 2132 outpatients with known ASCVD risk. In the group with unknown ASCVD risk, those predicted to have a 7.5% or higher 10-year MACE risk by CXR CVD-Risk had a higher adjusted hazard ratio (1.73) for MACE compared to those with lower risk. In the group with known ASCVD risk, CXR CVD-Risk predicted MACE beyond the traditional ASCVD risk score (adjusted HR, 1.88). The model showed improved discrimination and calibration compared to the traditional ASCVD risk score. The CXR CVD-Risk model identified a higher proportion of patients as statin-eligible compared to the traditional ASCVD risk score, particularly in those with unknown ASCVD risk. For example, 47.9% of patients with unknown ASCVD risk were identified as statin-eligible by CXR CVD-Risk, with 11.8% developing MACE, compared to 6.2% of those classified as ineligible. In the group with known ASCVD risk, 37.0% were identified as statin-eligible by CXR CVD-Risk, with 7.6% developing MACE, compared to 3.1% of those classified as ineligible. The study suggests that CXR CVD-Risk could help identify individuals at high risk for MACE who may benefit from primary prevention with statins, even when traditional risk factors are not available. The model could be implemented as an automatic tool in electronic medical records to opportunistically analyze existing CXRs, potentially improving statin eligibility decisions. However, the study has limitations, including its retrospective design and potential overestimation of event rates. The model may not be generalizable to racially and ethnically diverse populations. Despite these limitations, the results indicate that deep learning can provide valuable prognostic information from CXRs, complementing traditional risk assessment methods.