Machine Learning Models for Point-of-Care Diagnostics of Acute Kidney Injury

Abstract Background Computerized diagnostic algorithms could achieve early detection of acute kidney injury (AKI) only with available baseline serum creatinine (SCr). To tackle with this weakness, we tried to construct a machine learning model for AKI diagnosis based on point-of-care clinical features regardless of baseline SCr. Methods Patients with SCr > 1.3 mg/dL were recruited retrospectively from Wan Fang Hospital, Taipei. A Dataset A (n = 2,846) was used as the training dataset and a Dataset B (n = 1,331) was used as the testing dataset. Point-of-care features, including laboratory data and physical readings, were inputted into machine learning models. The repeated machine learning models randomly used 70% and 30% of Dataset A as training dataset and testing dataset for 1,000 rounds, respectively. The single machine learning models used Dataset A as training dataset and Dataset B as testing dataset. A computerized algorithm for AKI diagnosis based on 1.5x increase in SCr and clinician’s AKI diagnosis compared to machine learning models. Results The repeated machine learning models showed accuracy of 0.65 to 0.69. The single machine learning models showed accuracy of 0.53 to 0.74. The computerized algorithm show accuracy of 0.86 to 0.95. Clinician’s diagnosis showed accuracy of 0.52 to 0.57. The clinical features with leading impact on model output included blood lymphocyte, white blood cell, platelet, SCr, aspartate aminotransferase, systolic blood pressure, and pulse rate. Conclusions The machine learning models were able to diagnose AKI in the context of absent baseline SCr and showed superior accuracy than clinicians have.