Abstract
Background Early detection of acute kidney injury (AKI) is important for safe clinical practice. NHS England is implementing a nationwide automated AKI detection system based on changes in blood creatinine. Little has been reported on the similarities and differences of AKI patients detected by this algorithm and other definitions of AKI in the literature.
Methods We assessed the NHS England AKI algorithm and other definitions using routine biochemistry in our own health authority in Scotland in 2003 (adult population 438 332). Linked hospital episode codes (ICD-10) were used to identify patients where AKI was a major clinical diagnosis. We compared how well the algorithm detected this subset of AKI patients in comparison to other definitions of AKI. We also evaluated the potential ‘alert burden’ from using the NHS England algorithm in comparison to other AKI definitions.
Results Of 127 851 patients with at least one blood test in 2003, the NHS England AKI algorithm identified 5565 patients. The combined NHS England algorithm criteria detected 91.2% (87.6–94.0) of patients who had an ICD-10 AKI code and this was better than any individual AKI definition. Some of those not captured could be identified by algorithm modifications to identify AKI in retrospect after recovery, but this would not be practical in real-time. Any modifications also increased the number of alerted patients (2-fold in the most sensitive model).
Conclusions The NHS England AKI algorithm performs well as a diagnostic adjunct in clinical practice. In those without baseline data, AKI may only be seen in biochemistry in retrospect, therefore proactive clinical care remains essential. An alternative algorithm could increase the diagnostic sensitivity, but this would also produce a much greater burden of patient alerts.
Methods We assessed the NHS England AKI algorithm and other definitions using routine biochemistry in our own health authority in Scotland in 2003 (adult population 438 332). Linked hospital episode codes (ICD-10) were used to identify patients where AKI was a major clinical diagnosis. We compared how well the algorithm detected this subset of AKI patients in comparison to other definitions of AKI. We also evaluated the potential ‘alert burden’ from using the NHS England algorithm in comparison to other AKI definitions.
Results Of 127 851 patients with at least one blood test in 2003, the NHS England AKI algorithm identified 5565 patients. The combined NHS England algorithm criteria detected 91.2% (87.6–94.0) of patients who had an ICD-10 AKI code and this was better than any individual AKI definition. Some of those not captured could be identified by algorithm modifications to identify AKI in retrospect after recovery, but this would not be practical in real-time. Any modifications also increased the number of alerted patients (2-fold in the most sensitive model).
Conclusions The NHS England AKI algorithm performs well as a diagnostic adjunct in clinical practice. In those without baseline data, AKI may only be seen in biochemistry in retrospect, therefore proactive clinical care remains essential. An alternative algorithm could increase the diagnostic sensitivity, but this would also produce a much greater burden of patient alerts.
Original language | English |
---|---|
Pages (from-to) | 1853-1861 |
Number of pages | 9 |
Journal | Nephrology Dialysis Transplantation |
Volume | 30 |
Issue number | 11 |
Early online date | 28 Apr 2015 |
DOIs | |
Publication status | Published - Nov 2015 |
Bibliographical note
ACKNOWLEDGEMENTSWe acknowledge the data management support of the Grampian Data Safe Haven (DaSH) and the associated financial support of NHS Research Scotland, through NHS Grampian investment in the Grampian DaSH.
FUNDING
This work was funded through a personal fellowship for S.S. supported by the Wellcome Trust (reference number 102729/Z/13/Z).
Keywords
- acute kidney injury
- diagnosis
- epidemiology
- screening
Fingerprint
Dive into the research topics of 'Acute kidney injury – how does automated detection perform?'. Together they form a unique fingerprint.Datasets
-
Grampian Laboratory Outcomes Morbidity and Mortality Study (GLOMMS)
Black, C. (Creator) & Marks, A. (Creator), Grampian Data Safe Haven, 2014
http://www.abdn.ac.uk/ims/research/immunology/renal-304.php
Dataset
Profiles
-
Corri Black
- School of Medicine, Medical Sciences & Nutrition, Aberdeen Centre for Health Data Science
- School of Medicine, Medical Sciences & Nutrition, Applied Health Sciences - Personal Chair (Clinical)
- School of Medicine, Medical Sciences & Nutrition, Grampian Data Safe Haven (DaSH)
- School of Medicine, Medical Sciences & Nutrition, Chronic Disease Research Group
- School of Medicine, Medical Sciences & Nutrition, Farr Aberdeen
Person: Clinical Academic
-
Simon Sawhney
- School of Medicine, Medical Sciences & Nutrition, Applied Health Sciences - Senior Clinical Lecturer
- School of Medicine, Medical Sciences & Nutrition, Aberdeen Centre for Health Data Science
- School of Medicine, Medical Sciences & Nutrition, Farr Aberdeen
- School of Medicine, Medical Sciences & Nutrition, Grampian Data Safe Haven (DaSH)
Person: Clinical Academic