Machine learning models in electronic health records can outperform conventional survival models for predicting patient mortality in coronary artery disease