Under-Recognition of Kidney Disease in Cause-of-Death Certification: An Adjudication-Based Study With Development of an Exploratory Prediction Model.

Background Accurate determination of the cause of death (COD) is essential for estimating disease burden and guiding healthcare planning. The primary objective was to quantify discrepancies between death certificate documentation and adjudicated kidney disease-related deaths. The secondary objective was to develop and internally validate an exploratory prediction model for adjudicated kidney disease-related mortality. Methods A retrospective observational study was conducted in the medical intensive care unit (MICU) of a tertiary-care hospital in India. Medical records of 365 randomly selected deceased patients were reviewed. Kidney disease was defined to include acute kidney injury (AKI), chronic kidney disease (CKD), and acute-on-CKD, per KDIGO criteria. Candidate predictors were screened using Information Value (IV) and Variance Inflation Factor (VIF) analyses, followed by multivariable logistic regression. The model was developed in a 75% training cohort and validated in the remaining 25%. Performance was assessed using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUROC). Odds ratios (ORs) with 95% confidence intervals (CIs) were reported, and model coefficients were used to develop a risk calculator for predicting kidney disease-related death. Results Kidney disease was identified as contributing to death in 67.4% of cases following detailed chart review, compared with 45.2% documented on death certificates, indicating substantial underreporting. The present model identified oliguria (OR 20.48, 95% CI 8.02-52.31; p < 0.001) and hemodialysis (OR 13.46, 95% CI 2.65-68.47; p = 0.002) as the strongest predictors of kidney disease-related death, whereas hypotension, total protein, and cancer demonstrated borderline associations. The model demonstrated good predictive performance, achieving accuracies of 84.6% and 72.8% in the training and validation datasets, respectively. Discriminatory ability was excellent, with AUROC values of 0.905 in the training cohort and 0.856 in the validation cohort, indicating good model stability and generalizability. Conclusions COD attributable to kidney disease appears to be substantially underreported in death certification. A simple clinical prediction model demonstrated promising discriminatory performance in this exploratory analysis and may support more accurate attribution of kidney disease-related mortality. External validation in diverse healthcare settings is required before wider implementation.