The initial development and assessment of an automatic alert warning of acute kidney injury

Background. Acute kidney injury (AKI) recognition by clinicians has been shown to be suboptimal. Little work has focused on the use of an automated warning of a rise in a patient's creatinine, indicating AKI. Methods. Over 3 months in 2008 we ran a prospective observational study of 'alerts' sent by our Integrated Clinical Environment pathology system, identifying all patients with a >= 75% rise in their creatinine from its previous value. Information was collected on subsequent renal function, comorbidities and other potential predictors of survival. Results. In the 3-month period 463 adults with a first episode of AKI were identified by an alert; 87% were hospital inpatients. Median follow-up was 404 days. In-hospital mortality was 36% for those who were admitted. After performing Weibull survival analysis, significant predictors of poorer survival were the presence of metastatic, haematological or lower risk malignancy, a residential or nursing home address and higher age, number of non-malignant comorbidities or C-reactive protein level. Predictors of better survival were higher serum albumin level or nadir GFR during the episode and Indian subcontinent ethnicity. A receiver-operator curve for a prognostic score developed from the analysis showed an area under the curve of 0.84. Conclusions. The alerts identified a group of AKI patients who are at moderately high risk of death. The prognostic score using a combination of covariates shows early promise. Both the alerts and the score warrant further development as tools for earlier intervention in AKI.