Automated Detection of Acute Kidney Injury in Routine Healthcare

Sawhney, Simon, Marks, Angharad, Fluck, Nick, Prescott, Gordon orcid iconORCID: 0000-0002-9156-2361, Simpson, William G., Tomlinson, Laurie and Black, Corri (2015) Automated Detection of Acute Kidney Injury in Routine Healthcare. Nephrology Dialysis Transplantation, 30 (3). pp. 441-442. ISSN 0931-0509

Full text not available from this repository.

Official URL:


Introduction and Aims: Early detection of acute kidney injury (AKI) is a cornerstone of safe clinical practice. NHS England is introducing a nationwide automated AKI detection system based on changes in blood creatinine. Its diagnostic accuracy and impact on clinical practice have not previously been reported. The challenge lies in identifying AKI early but without misclassifying patients who do not have AKI.

Methods: We assessed the NHS AKI algorithm using routine biochemistry in a single health authority in Scotland in 2003 (adult population 438,332).We used hospital episode statistics (HES) coded AKI as reference standard for clinically important events. We evaluated the impact of modifications to the algorithm based on previously reported AKI definitions.

Results: Of 127,851 patients with at least one blood test in 2003, the NHS AKI algorithm identified 5565 patients. The algorithm correctly captured 91.2% (87.6-94.0) of HES AKI patients but missed 8.8% of patients. Missed patients had less diabetes and cardiac disease, but many could be identified by modifying algorithm criteria. Any modifications, however, increased the number of alerted patients (twofold in the most sensitive model).

Conclusions: AKI can be identified using an automated algorithm, but patients may be missed during real-time data collection. Small changes to the algorithm improved its performance, but involved twice as many patient alerts. Care must therefore be taken to avoid misdiagnosis when using the algorithm in clinical practice or research.

Repository Staff Only: item control page