Predicting the Readmission of Heart Failure Patients through Data Analytics

Reducing the costs and improving the quality of treatment in hospital systems as well as demands for better treatment from patients in order to keep them away from readmissions are two main issues healthcare systems have faced. In order to solve such challenges, predicting the occurrence of re-hospitalisation with data mining techniques would be so worthwhile. In this study, we are seeking to predict the occurrence of re-hospitalisation of the heart failure patients in two time-horizons (1-month and 3-month) via deployment of classification algorithms (i.e. decision trees, artificial neural networks, support vector machines and logistic regression). Two criterions (as main criterions) such as AUC (area under curve) and ACC (accuracy) have been calculated and assessed for classifying the prediction-power of the models in each time-horizon (outcome/target). We also have calculated some other criterions such as recall, precision and F1-Score. Then, we identified the importance and contribution of the variables for each outcome. Therefore, the variables whose contribution/importance changes over time are differentiated. It is noteworthy to say that this study is done under the scrutiny of an expert cardiologist. Trained nurses and expert cardiologist monitored the dataset every day, which was a hard and valuable measure to conduct. Finally, the dataset does not have missing values and noises. This research can be the basis for prospective medical studies and projects.