AI model predicts kidney disease with 99% accuracy in AP’s Uddanam

Hyderabad: An artificial intelligence model trained on patient data from the Uddanam region of Andhra Pradesh has predicted chronic kidney disease (CKD) with nearly 99% accuracy, opening the door for much earlier detection of the largely ‘silent' illness in one of the high risk hotspots.The findings were published in ‘Scientific Reports' in a study titled ‘Stacking ensemble model for predicting chronic kidney disease in the Uddanam region of India with unknown etiology.' CKD is described in the paper as a major global public health challenge, difficult to detect in its early stages because it shows few obvious symptoms. The Uddanam coastal belt in Srikakulam district has emerged as a concentrated pocket of the disease. The region spans 30-40 km inland and covers five mandals — Ichchapuram, Kanchili, Kaviti, Sompeta and Vajrapu Kotturu — across nearly 120 villages. The area has reported unusually high rates of CKD of unknown cause, commonly known as Uddanam nephropathy or Uddanam CKD), often in people without traditional risk factors such as diabetes or hypertension. Despite multiple studies, the exact cause remains unclear. The paper outlines key hypotheses including environmental toxins, genetic susceptibility and occupational exposure. Earlier published work, summarised by the authors, highlights contaminated drinking water, intensive agrochemical use, genetic factors and chronic dehydration as recurring themes. This persistent uncertainty, the researchers argue, demands more precise analytical tools. Local data, rebuilt for AI For the study, the researchers created a new ‘Uddanam CKD' dataset using health records from hospitals and clinical centres in the region. The data underwent rigorous preprocessing to improve quality. Outliers were detected and removed, and extensive exploratory data analysis was conducted, including statistical reviews, feature inspection and validation of the CKD versus non-CKD target variable. The study was led by Rakesh Salakapuri of Symbiosis Institute of Technology, Hyderabad campus, and Panduranga Vittal Terlapu of Aditya Institute of Technology and Management, Tekkali, Srikakulam. The team then built a stacking ensemble machine learning model, combining multiple algorithms to work together rather than relying on a single technique. Principal component analysis (PCA) was applied to reduce data complexity before model training. The researchers report that the PCA-assisted stacked model outperformed existing state-of-the-art systems, delivering higher predictive accuracy and stronger generalisability across the dataset. What the data shows Analysis of patient characteristics showed that people with CKD were generally older, with mean and median ages of 52.1 and 55 years, respectively. Neutrophil counts ranged between 59.1-61.2 in CKD patients and 59.4-61.4 in non-CKD patients. Men made up the majority of both groups — 74.7% among CKD patients and 75.4% among those without CKD. Blood sugar emerged as a key differentiator, with CKD patients showing higher mean and median values of 180.29 and 156. Mean blood urea levels were also higher at 72.30. Comorbidities were common: 60.4% had diabetes, 46.0% had anaemia, 60.4% had hypertension and 29.1% had coronary artery disease.Statistically significant differences were also observed in lymphocytes, eosinophils, monocytes, serum urea, basophils, creatinine, bilirubin, uric acid and red blood cell counts — the signals the model learns to associate with kidney damage in the Uddanam population.