Predicting Readmissions in Diabetic Patients: A Dual-Scenario Approach for Hospital Resource Optimization

Diabetes is a global disease which is characterized by high levels of blood glucose, due to the inability of the body to produce insulin or the insufficient use of the insulin produced. Diabetes requires periodic check-ups, medicines, constant watch and follow ups. The idea of early flagging is crucial for effective distribution of patients within the healthcare organization. In the present analysis, a dataset containing data on inpatient encounters from 130 US hospitals is used for the purpose of readmissions prediction of diabetic patients. The research focuses on two distinct prediction scenarios, firstly ’Necessity Case Readmission’ for the better conditioning of the hospital resource management; and follwed by ’Criticality Case Readmission’ for the identification of patients with severe complications as early as possible. To increase the model performance, several data balancing methods were utilized and provided a balance between the classes. Filter techniques were employed to select the most important variables for prediction purposes. Machine learning models were assessed and the Optuna hyperparameters were applied to the best performing models. The proposed approach yielded a recall rate of 0.80, a F1-score of 0.78, and 80% accuracy. These predictions aim at effectively helping the hospital management provide better patient care with efficiency in the use of the scarce resources. The research paper suggests that both necessity and criticality predictions can be useful in developing a proactive strategy for the healthcare industry.