Differential Diagnosis of Heart Disease in Emergency Departments Using Decision Tree and Medical Knowledge

Generating differential diagnosis has been a subjective process primarily relying on a physician’s experience. However, the increased availability of electronic health records (EHRs) means that this process has the potential to benefit from machine learning-based decision support technology. No differential diagnosis models are currently available for heart disease, particularly for physicians in emergency departments (EDs). In this paper, we applied the decision tree method to automatically build a heart disease differential diagnosis model from structured and unstructured ED data. Our results show that the automatically learned model can achieve a classification accuracy of 89%. Our study demonstrates that data-driven differential diagnosis rules can be automatically learned from analyzing EHR data and that this learning can be clinically meaningful when merged with external medical knowledge.