CONTINUOUS PATIENT DATA INTEGRATION FOR EARLY KIDNEY-RISK IDENTIFICATION, LONGITUDINAL MONITORING AND PREDICTIVE HEALTH ANALYTICS - (KIDNEY SHIELD)

Abstract

Chronic kidney disease develops slowly and often without symptoms, so many patients are diagnosed only after a large part of their kidney function has already been lost. The information needed to notice the decline earlier usually exists, but it is scattered across laboratory reports, outpatient visit notes, prescription records, and home blood pressure readings that are rarely viewed together. This paper presents Kidney Shield, a data integration and analytics platform that continuously brings these patient records into one longitudinal store, tracks kidney function over time, and identifies patients whose risk of progression is rising while there is still time to intervene. The platform ingests data from the hospital laboratory information system, the electronic health record, the pharmacy system, and periodic uploads from home monitoring devices. An extract, transform, and load pipeline standardises units, maps test names to common codes, links records belonging to the same patient through a master patient index, and removes duplicate or implausible values. Serum creatinine results are converted into an estimated glomerular filtration rate using the CKD-EPI equation, and urine albumin to creatinine ratios are aligned with each patient's timeline so that kidney function can be followed as a continuous series rather than as isolated reports. On top of this longitudinal store, the analytics layer computes trend features such as the slope of eGFR over the last twelve months, the variability of blood pressure, the number of months with poorly controlled blood glucose, and exposure to medicines known to affect the kidney. A gradient boosting classifier estimates the probability that a patient will move to a more advanced stage of chronic kidney disease within the next two years, and a mixed-effects trajectory model projects the likely eGFR path. Rule-based checks run alongside the model to flag rapid declines that need immediate attention regardless of the predicted score.