Abstract
Background
National Health Services (NHS) England mandates that an acute kidney injury (AKI) detection algorithm be embedded in laboratories. We evaluated the implementation of the algorithm and the consistency of alerts submitted to the United Kingdom Renal Registry (UKRR).
Methods
Code was developed to simulate the syntax of the AKI detection algorithm, executed on data from local laboratories submitted to the UKRR, including alerts and serum creatinine (SCr) results spanning 15 months before and after the alert submission. Acute kidney injury alerts were categorized into stages 0/1/2/3. Inter-rater agreement (Gwet’s AC1) was used to compare local and centrally derived alerts at individual laboratory and commercial laboratory information management system (LIMS) levels, penalizing extreme disagreements.
Results
The analysis included 9,096,667 SCr results from 29 labs (475,634 patients; median age 72 years, 47% female) between algorithm activation and data extraction (September 30, 2020). Laboratories and the central simulation generated 1,579,633 and 1,646,850 non-zero AKI alerts, respectively. Agreement was high within known laboratory information management system providers (0.97–0.98) but varied across individual laboratories (overall range 0.17–0.98, 0.17–0.23 in three). Agreement tended to be lower (Gwet’s AC1 0.88) with the highest baseline SCr quartile (median 164 μmol/L).
Conclusions
Overall, alerts submitted to the UKRR are a valid source of AKI surveillance but there are concerns about inconsistent laboratory practices, incomplete adoption of the NHSE algorithm code, alert suppression, and variable interpretation of guidelines. Future efforts should audit and support laboratories with low agreement rates, and explore reasons for lower agreement in individuals with pre-existing CKD.
National Health Services (NHS) England mandates that an acute kidney injury (AKI) detection algorithm be embedded in laboratories. We evaluated the implementation of the algorithm and the consistency of alerts submitted to the United Kingdom Renal Registry (UKRR).
Methods
Code was developed to simulate the syntax of the AKI detection algorithm, executed on data from local laboratories submitted to the UKRR, including alerts and serum creatinine (SCr) results spanning 15 months before and after the alert submission. Acute kidney injury alerts were categorized into stages 0/1/2/3. Inter-rater agreement (Gwet’s AC1) was used to compare local and centrally derived alerts at individual laboratory and commercial laboratory information management system (LIMS) levels, penalizing extreme disagreements.
Results
The analysis included 9,096,667 SCr results from 29 labs (475,634 patients; median age 72 years, 47% female) between algorithm activation and data extraction (September 30, 2020). Laboratories and the central simulation generated 1,579,633 and 1,646,850 non-zero AKI alerts, respectively. Agreement was high within known laboratory information management system providers (0.97–0.98) but varied across individual laboratories (overall range 0.17–0.98, 0.17–0.23 in three). Agreement tended to be lower (Gwet’s AC1 0.88) with the highest baseline SCr quartile (median 164 μmol/L).
Conclusions
Overall, alerts submitted to the UKRR are a valid source of AKI surveillance but there are concerns about inconsistent laboratory practices, incomplete adoption of the NHSE algorithm code, alert suppression, and variable interpretation of guidelines. Future efforts should audit and support laboratories with low agreement rates, and explore reasons for lower agreement in individuals with pre-existing CKD.
| Original language | English |
|---|---|
| Number of pages | 9 |
| Journal | Journal of Nephrology |
| Early online date | 4 Aug 2024 |
| DOIs | |
| Publication status | E-pub ahead of print - 4 Aug 2024 |
Bibliographical note
Publisher Copyright:© The Author(s) 2024.