AI Model Achieves 99% Accuracy in Predicting Kidney Disease

An innovative artificial intelligence model has demonstrated nearly 99% accuracy in predicting chronic kidney disease (CKD) in the Uddanam region of Andhra Pradesh. This advancement could lead to earlier detection of CKD, a condition often termed a “silent illness” due to its elusive early symptoms. The study detailing these findings was published in Scientific Reports and is titled “Stacking ensemble model for predicting chronic kidney disease in the Uddanam region of India with unknown etiology.”

The Uddanam coastal area, located in the Srikakulam district, has been identified as a high-risk zone for CKD, specifically known as Uddanam nephropathy. This condition disproportionately affects individuals without traditional risk factors, such as diabetes or hypertension. The region encompasses approximately 120 villages across five mandals: Ichchapuram, Kanchili, Kaviti, Sompeta, and Vajrapu Kotturu, and stretches 30-40 km inland.

Despite numerous studies, the precise cause of CKD in this area remains undetermined. The researchers outlined several hypotheses, including environmental toxins, genetic predispositions, and occupational exposures. Previous research has pointed to factors such as contaminated drinking water, intensive use of agrochemicals, genetic influences, and chronic dehydration as potential contributors to the disease.

Developing a Local Dataset for AI Analysis

To conduct this study, the team built a unique dataset focused on Uddanam CKD, utilizing health records from local hospitals and clinics. The data underwent comprehensive preprocessing to enhance its quality. Outliers were identified and removed, and an extensive exploratory data analysis was performed, which included statistical assessments and validation of CKD versus non-CKD classifications.

Led by Rakesh Salakapuri from the Symbiosis Institute of Technology in Hyderabad, and Panduranga Vittal Terlapu from the Aditya Institute of Technology and Management in Tekkali, the research team applied a stacking ensemble machine learning model. This approach merges multiple algorithms to improve predictive performance rather than relying on a single method. Additionally, principal component analysis (PCA) was utilized to simplify the data before training the model. The result was a PCA-assisted stacked model that surpassed existing systems in predictive accuracy and generalisability.

Insights from Patient Data

The analysis revealed notable trends among CKD patients. The average age of those affected was 52.1 years, with a median age of 55 years. Blood sugar levels were significantly higher in CKD patients, averaging 180.29. The mean blood urea levels for these patients were recorded at 72.30.

Comorbidities were prevalent, with 60.4% of CKD patients also having diabetes, and similar rates for hypertension and anaemia. The demographic breakdown indicated a higher prevalence of CKD among men, comprising 74.7% of the CKD group and 75.4% of the non-CKD group. The study further highlighted statistically significant differences in various blood parameters, including lymphocyte and creatinine levels, which the model identifies as indicators of kidney damage within the Uddanam population.

This groundbreaking research not only enhances understanding of CKD in a vulnerable area but also showcases the potential of AI applications in healthcare. With further exploration and validation, the AI model could become a crucial tool in combating chronic kidney disease, ultimately improving patient outcomes in high-risk regions.