Advancing Healthcare Outcomes with AI: Predicting Hospital Readmissions in the USA

Abstract

The issue of readmission rates in hospitals has been described as both a serious and perplexing problem in America's healthcare system. The high persistence of readmission rates underscores the urgent need for improvement in better tools and techniques for the forecasting and management of occurrences with efficiency.  The chief objective of this research was to devise and ameliorate AI models that can effectively predict patient readmissions. Through machine learning and data analytics, this study worked toward developing tools that will highlight patients at a high risk of readmission, which can be targeted with interventions by healthcare providers. The hospital readmission dataset used in this study comprised a comprehensive collection of patient-related data aimed at understanding and predicting readmissions.  The dataset was thereby developed using electronic health records which capture all clinical activities - diagnosis code treatment history, results of labs, and medication-related prescriptions. Demographic details related to patients will include: age, sex, and ethnic background - for contextualizing at the population level. This clinical information was complemented by unstructured data, such as clinical notes that give further detailed insight into patient conditions and advice on follow-up care. Several models were considered for classification tasks such as Random Forest Classifier, Logistic Regression, and XG-Boost Classifier. Some of the key metrics used to quantify the model's effectiveness included accuracy, precision, recall, F1-score, and ROC-AUC. Gradient Boosting had the highest scores on all four metrics and maximum accuracy and F1-score, showing the best all-rounded performance in prediction. Interpreting healthcare model outputs provides insightful predictions to inform clinical decisions. Care strategies have to be developed based on predictive insights and patient segmentation analysis to enhance the outcomes of patients. AI-driven insights will thus require a strategic approach to the integration of AI-driven models in the functioning of the hospital.