Early identification of patients with chronic kidney disease of uncertain etiology (CKDu) in Anuradhapura district of Sri Lanka using data science algorithms.

BACKGROUND: Over the past two decades, there has been a noticeable emergence of chronic kidney disease of uncertain etiology (CKDu) as a substantial contributor to the burden of chronic kidney disease (CKD) in rural Sri Lanka where early CKDu patients have less symptoms. This study was conducted to develop data science models to identify whether a selected individual is healthy or suffering from CKD and to determine the severity of patients with CKDu. METHODS: Seventeen attributes were considered for 276 individuals and Logistic Regression, Feed-Forward Neural Network (FFNN), Probabilistic Neural Network (PNN), and Support Vector Machine were employed to build the models. RESULTS: In the first phase, participants were classified as CKDu and healthy. In the second phase, they were characterized by severity as having microalbuminuria or macroalbuminuria. CKDu was linked to habitation, water source, age, and past hantavirus exposure in the first phase. Six factors contributed to severe CKDu in the second phase with two further factors: years of CKD and daily water consumption. The FFNN and PNN had 87.8% accuracy when identifying CKDu patients while FFNN outperformed PNN with a greater F1 score. The FFNN outperformed other models in detecting the severity with 74.36% precision. CONCLUSION: The Feed-Forward Neural Network was the better model for identifying patients in advance with kidney diseases and predicting the severity of CKDu. These models will be helpful to clinicians as a decision-support tool in identifying patients with CKDu and to determine the severity in advance to provide appropriate therapy and measures to minimize the progression.