Understanding the relationship between healthcare processes and in-hospital weekend mortality using MIMIC III

Abstract

Abstract Studies showed evidence that hospitals that offer services 24 h seven days a week have a higher rate of mortality on weekends. However, there is limited information related to the type of hospital admissions effected, and the factors responsible for this phenomenon. This study proposes a novel approach to understand the factors that influence hospital mortality rate in patients admitted under Elective, Emergency and Urgent Care units. First, we compare the in-hospital mortality rate of patients admitted to the intensive care unit on weekdays and weekends. Then we extract the features from the MIMIC III database that support the prediction. The extracted features were then used to train and test four machine learning models: logistic regression, decision tree, gradient boosted tree and random forest. These statistical models help to understand the significant factors that influence weekend mortality predictions. Model performance is demonstrated based on AUC, kappa, precision and recall values. The results from the analysis demonstrate that patients admitted to elective care on the weekend have a high risk of mortality compared to weekday admissions. Our analysis also indicates that the reduced availability of caregivers and the number of lab tests performed are significant contributors to this phenomenon.