Computerized algorithms compared with a nephrologist's diagnosis of acute kidney injury in the emergency department.

• Published in: European Journal of Internal Medicine
• Main Authors: Jonsson, Arnar Jan, Kristjansdottir, Ingibjorg, Lund, Sigrun Helga, Palsson, Runolfur, Indridason, Olafur S
• Other Authors: 1 University of Iceland, Reykjavik, Iceland; Internal Medicine Services, Landspitali - The National University Hospital of Iceland, Reykjavik, Iceland. 2 Internal Medicine Services, Landspitali - The National University Hospital of Iceland, Reykjavik, Iceland. 3 University of Iceland, Reykjavik, Iceland. 4 University of Iceland, Reykjavik, Iceland; Division of Nephrology, Landspitali - The National University Hospital of Iceland, Reykjavik, Iceland. 5 Division of Nephrology, Landspitali - The National University Hospital of Iceland, Reykjavik, Iceland. Electronic address: [e-mail].
• Format: Article in Journal/Newspaper
• Language: English
• Published: Elsevier Science 2019
• Subjects: Acute kidney injury; Computer algorithms; Diagnosis; KDIGO criteria; Serum creatinine; Nýrnabilun; Reiknirit; Algorithms; Iceland
• Online Access: http://hdl.handle.net/2336/620876; https://doi.org/10.1016/j.ejim.2018.11.013

To access publisher's full text version of this article click on the hyperlink below The aim of this study was to examine acute kidney injury (AKI) diagnosis based on different computerized algorithms compared with a nephrologist's diagnosis in patients visiting an emergency department (ED) of a university hospital. In this retrospective study, we used electronic medical records at the University Hospital in Reykjavik to identify all patients aged ≥18 years, who presented to the ED in the year 2010 with an elevated serum creatinine (SCr) level. All SCr values were reviewed and a nephrologist determined whether AKI was present using the KDIGO SCr criteria and clinical data. Computerized algorithms based on the KDIGO SCr criteria, accounting for various time intervals for baseline SCr and changes in follow-up SCr, were constructed using the statiscal software R. At 53,816 ED visits, SCr was measured in 15,588 patients for a total of 21,559 measurements. Elevated SCr was observed in 2878 (18.4%) patients. Strict adherence to the KDIGO SCr criteria yielded a 79% sensitivity, 94% specificity, 68% positive predictive value (PPV) and 96% negative predictive value (NPV) for the diagnosis of AKI. Allowing for a longer time frame (>365 days) for baseline SCr, resulted in 93% sensitivity, 96% specificity, 80% PPV and 99% NPV. The algorithms which included a decrease in SCr from the index ED value yielded a sensitivity of 97% but lower specificity, 74% and 80%. The algorithms that perform best yield excellent sensitivity and specificity and could be used to identify patients with AKI in the ED to enhance early diagnosis and treatment. Landspitali University Hospital Science Fund, Reykjavik, Iceland