Predicting Hospital Readmission Rates Using Data Mining Techniques

Introduction: Due to the high costs of patient hospitalization, the increasing demand for readmission has led to a lack of suitable services provided by hospitals. The present study aimed to predict the risk of readmission among patients admitted to a large subspecialty hospital in Tehran using data mining techniques. Method: This retrospective cohort study employed data mining techniques following the CRISP-DM methodology to identify factors contributing to patient readmission. The study analyzed 47,892 electronic medical records from a large public hospital in Tehran, analyzing data from August 2018 to August 2019. The study utilized demographic and clinical data, extracting patterns to provide insights into readmission risks and support healthcare decision-making. Key algorithms included neural networks and the C5 decision tree. Model evaluation was measured by the accuracy rate in predicting readmission. Results: The findings from the neural network analysis revealed that the type of discharge, inpatient department, and length of stay significantly impacted readmission rates, with coefficients of 0.28, 0.21, and 0.16, respectively. The neural network model achieved an accuracy rate of 61.2% in predicting readmission. The analysis using the C5 decision tree algorithm showed that the length of stay, number of medications prescribed, and type of discharge had the most influence on readmission rates, with coefficients of 0.12, 0.11, and 0.10, respectively. Among the 37,832 patients analyzed, 11.95% experienced readmission, with 8.63% readmitted once, 2.32% twice, and 1% three or more times. Non-emergency admissions, non-surgical treatments, and specific discharge types were notable factors in readmission rates. Conclusion: Findings revealed that key factors, including the type of discharge, inpatient department, length of stay, and the number of medications prescribed, have substantial impacts on readmission likelihood. The data mining models achieved a suitable accuracy to predict readmissions as well as highlighted variables related to readmissions. Implementing strategies to use data mining models with better data management and quality can provide actionable insights for decision-making on patients' risk of readmission.