Predictors of In-Hospital Length of Stay among Cardiac Patients: A Machine Learning Approach

Purpose: Cardiovascular diseases remain a major cause of morbidity and mortality and are currently a significant strain on the healthcare system.Knowledge of the predictors of LoS is important to streamline care, resource utilization and the outcomes of the patients.The purpose of this study was to determine socio-demographic factors that could be used to predict LoS among patients with heart-related conditions in the Greater Accra Region in Ghana using machine learning methods.Methods: A retrospective study was done through the hospital records of patients with heart related diseases admitted between January 1 st 2022 and December 31 st , 2023 from the District Health Information Management System (DHIMS2).Three predictive models, the Decision Tree, Random Forest, and the Linear Regression, were used to assess the determinants of the LoS.Results: The average LoS was 5.5 days.The highest prevalence of diagnoses was hypertensive heart disease (56%) and heart failure (31%) in both genders.Age, educational attainment, and occupational status were found important determinants of LoS.Tertiary-educated and formally employed patients had the longest admissions.Interaction effects showed lower LoS in younger employed patients, and highly educated in the informal economy.The decision tree model yielded the best prediction (R 2 = 37.97%).Conclusion: Decision trees may serve as an effective machine learning model to point at the complicated and non-linear predictors of LoS of cardiac patients.These results stress the significance of developing a resource planning and discharge plan based on the socio-demographical characteristics of the patients.