The multi-view deep discharge risk assessment model achieved 96.5% accuracy, 94.1% precision, 72.7% recall, 82.0% F1 score, and 85.5% AUC in elderly CHD patients.
Does a multi-view deep discharge risk assessment model (MDDRA) accurately predict post-discharge risk in elderly patients with coronary heart disease?
A novel multi-view deep learning model (MDDRA) demonstrated high accuracy and predictive performance for assessing post-discharge risk in elderly patients with coronary heart disease.
Absolute Event Rate: 0% vs 0%
Abstract Background The discharge assessment of elderly patients with coronary heart disease (CHD) is very important for their discharge timing and risk prediction. The current disease risk assessment has gradually shifted from the traditional statistical analysis to the use of machine learning technology to deal with clinical multi-dimensional, multi-source, heterogeneous and fluctuating data. Purpose The discharge risk prediction model was constructed by machine learning to assist the discharge assessment of elderly patients with CHD. Methods Based on the clinical data of 1007 subjects selected from the elderly coronary heart disease cohort of our Hospital and Friendship Hospital, 148 key features significantly associated with post-discharge risk assessment were selected to construct a comprehensive dataset (Including the whole process from admission to follow-up). According to the prognosis of patients after discharge, they were divided into no risk and at risk (the ratio was 8:1), and the feature fusion of patients was carried out with double-layer attention fusion network. The database is divided into the training set and the test set in the ratio of 8:2, and the five-fold cross-validation method is used for training. Finally, the model was evaluated using five evaluation indicators: accuracy (Acc), precision (P), recall (R), F1 score (F), and area under ROC curve (AUC). Four deep learning models (BiLSTM, CNN, Gate Recurrent Unit (GRU), Transformer Encoder and Random Forest (RF)) were used for comparison experiments. Results We built the multi-view deep discharge risk assessment model (MDDRA) through machine learning. After gradual optimization, the prediction accuracy of the model reached 96.5%, precision reached 94.1%, recall rate reached 72.7%, F1 score reached 82.0%, and AUC reached 85.5%. Compared with the other four deep learning models, this model has more advantages in the discharge assessment of elderly CHD. On the basis of building the model, we will further correspond the results of machine learning to clinical medical interpretation, enhance its interpretability, and better help clinicians to judge the prognosis of elderly patients with CHD. Conclusion The MDDRA model built by combining clinical and artificial intelligence has certain advantages in assessing discharge risk of elderly patients with CHD.
Yan et al. (Sat,) reported a other. The multi-view deep discharge risk assessment model achieved 96.5% accuracy, 94.1% precision, 72.7% recall, 82.0% F1 score, and 85.5% AUC in elderly CHD patients.