Prediction of Length of Stay for Heart Disease Patients using Ensemble Machine Learning Approach

Heart disease is a leading cause of death worldwide and requires immediate medical attention. However, prolonged and ineffective treatment can be detrimental to patients. There is a standard length of stay (LOS) for treating inpatients with certain diseases, including heart disease. Hospitals have limited resources that must be effectively managed to meet the demands of health services. Therefore, it is essential to accurately predict the LOS of heart disease patients, particularly for managing inpatient beds. In this study, we propose ensemble machine learning approaches i.e., the random forest (RF) method and extreme gradient boosting (XGBoost). Logistic regression is used as a baseline for experiments. The proposed models are applied to 15,757 data points obtained from 2017 to 2019 for the patients of Dayanand Medical College and Hospital, Ludhiana, Punjab, India. The experimental results show that the RF method achieves the highest F1 score of 83.8%. The features that contribute the most to the LOS include the category from demographics, risk factors, and laboratory test results. Hence, the healthcare should focus on predominant risk factors, i.e. hypertensions and other significant features in the laboratory results to effectively treat the patients.

Submitted to: Medical & Biological Engineering & Computing Journal