Multi-stage Chronic Kidney Disease Classification on Longitudinal Data

Chronic Kidney Disease (CKD) presents a significant global health challenge, often going unnoticed in patients until reaching advanced stages. Late-stage CKD profoundly impacts patients’ lifestyles. It often necessitates weekly dialysis or kidney transplants, which both require costly medical support. Detecting early-stage CKD, however, facilitates preventive measures through lifestyle changes and medical interventions. This highlights the importance of early detection and accurate staging. Recent advancements in machine learning offer immense promise for diagnosing and identifying the CKD stages. However, most studies focus on only binary classification (CKD or not CKD) using cross-sectional data. Nonetheless, it is often observed that longitudinal analysis is more suitable for long-term disease prediction, leveraging extensive temporal data. In this study, we conducted an analysis using a comprehensive dataset of blood test results obtained from the Welsh Results Reports Service (WRRS) accessed through the Secure Anonymised Information Linkage (SAIL) Databank. By utilizing longitudinal dataset and employing machine learning techniques, namely Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (Bi-LSTM) algorithms, we present the first study to classify all five stages (from 1st to 5th) of CKD, including both early and late stages. These techniques enabled us to determine the stages of CKD in patients with precision. We also compare our models against cross-sectional techniques commonly used in the literature, namely RF, SVM, Decision Tree and Logistic Regression. Our findings indicate that the longitudinal model yields better results. This could potentially be valuable for General Practitioners (GPs) in identifying CKD early for referral.