Background
Electronic medical records (EMRs) can be used to accurately predict the occurrence of a variety of diseases, which can contribute to precise medical therapies. While there is a growing interest in the application of machine learning (ML) techniques to solve clinical problems, the use of deep-learning in healthcare has just gained attention recently. Deep Learning, such as deep neural network (DNN), has achieved impressive results in the areas of speech recognition, computer vision, and natural language processing in recent years. However, deep learning was difficult to understand due to the complexities in its framework. Furthermore, this method had not demonstrated a better performance comparing to other conventional ML algorithms. It needs to show the performance of DNN relative to other approaches before the method can take a place in the healthcare systems.

Methods
We utilized a large population-based EMR database of around 800,000 patients to compare DNN with 3 other ML approaches for predicting 5-year stroke occurrence in Taiwan population.

Results
The DNN and gradient boosting decision tree (GBDT) can results in a better performance compared to logistic regression (LR) and support vector machine (SVM) approach. Meanwhile, DNN can achieve optimal results by using smaller amounts of patient data when comparing to the GBDT method. Algorithms based on DNN and GBDT can achieve high area under the curve values (0.915 and 0.918) for prediction of further stroke occurrence within 5 years.

Conclusions
In this work, we showed that DNN is a good predictive method in manages of EMRs. Further research is necessary to determine the feasibility of applying DNN in the clinical setting and to determine whether use of DNN could lead to improved clinical care and patients’ outcomes.