Electronic Medical Record-Based Predictive Model for Acute Kidney Injury in an Acute Care Hospital

Abstract

Patients with acute kidney injury (AKI) are at risk for increased morbidity and mortality. Lack of specific treatment has meant that efforts have focused on early diagnosis and timely treatment. Advanced algorithms for clinical assistance including AKI prediction models have potential to provide accurate risk estimates. In this project, we aim to provide a clinical decision supporting system (CDSS) based on a self-learning predictive model for AKI in patients of an acute care hospital. Data of all in-patient episodes in adults admitted will be analysed using "data mining" techniques to build a prediction model. The subsequent machine-learning process including two algorithms for data stream and concept drift will refine the predictive ability of the model. Simulation studies on the model will be used to quantify the expected impact of several scenarios of change in factors that influence AKI incidence. The proposed dynamic CDSS will apply to future in-hospital AKI surveillance in clinical practice.