Predictive Modeling of Kuwaiti Chronic Kidney Diseases (KCKD): Leveraging Electronic Health Records for Clinical Decision-Making

Abstract

Chronic kidney disease (CDK) represents a significant public health concern globally, and its prevalence is on the rise. In the context of Kuwait, this study addresses the imperative of predicting CKD by leveraging the wealth of information embedded in electronic health records (EHRs). The primary objective is to develop a predictive model capable of early identification of individuals at risk for CKD, thereby enabling timely interventions and personalized healthcare strategies and equip clinicians with information that enhances their ability to make well-informed decisions regarding prognoses or therapeutic interventions. In this study, a dataset has been created from Kuwaiti healthcare institutions, emphasizing the richness and diversity of patient information encapsulated in EHRs and a feature engineering step has been applied for labeling it. Various ensemble learning algorithms, Ada Boost, Extreme Gradient Boosting, Extra Trees, Gradient Boosting, Random Forest, and various single learning algorithms, Decision Tree, K-Nearest Neighbors, Logistic Regression, Multilayer Perceptron, Stochastic Gradient Descent, Support Vector Machines, have been implemented. By examining the empirical findings of our tests, our results showcase the models’ capability to identify individuals at risk for CKD at an early stage, facilitating targeted healthcare interventions. Decision Tree was the best classifier achieving 99.5% accuracy and 99.3% macro averaged f1-score.