Evaluation of Features for Prediction of Hospitalization in Patients with End Stage Renal Disease

Abstract: End Stage Renal Disease (ESRD) is a chronic kidney disease which results in lifelong care and regular dialysis treatments. It is both an expensive and extensive diagnosis that causes a lot of suffering for the patients. The workload on the healthcare system is increasing drastically, due to a population that lives longer with more chronic diseases. In this project, input data to an AI system was investigated and changed. The reviewed system predicts whether patients will be hospitalized in the near future or not. The input data consisted of features based on patients’ medical histories and the purpose of this project was to manually create features with new or modified information. This was done in order to improve the system’s performance with help of experiences from healthcare professionals. A test environment was used to evaluate the performance of the new feature sets. The newly created feature sets resulted in better predictions in some cases and worse in others compared to the original feature set. The results were measured with a Z-test with 95% confidence interval and the features were compared through a feature importance method. This project has shown that it is possible to improve AI predictions with the use of experiences from healthcare professionals during feature extraction. Since the improvements were just slightly better, the conclusion from this project was that larger structural changes are probably needed in order to convincingly increase the AI system’s prediction performance.