DISCRIMINATIVE ABILITY OF A RISK SCORE IN PREDICTING THE DEVELOPMENT OF ACUTE KIDNEY INJURY IN THE EMERGENCY DEPARTMENT

Acute kidney injury (AKI) is a preventable, yet common syndrome in patients presenting to the emergency department and is associated with a higher morbidity and mortality, particularly in low resource settings. We previously demonstrated extremely poor outcomes in critically ill patients with AKI in our setting. 0by25 was a bold initiative aimed at reducing preventable AKI deaths through early recognition and management of patients at increased risk of AKI including using a symptom-based risk score. The current study aimed to assess the predictive ability of an AKI risk score previously deployed in Malawi to predict AKI in sick patients presenting to the emergency department at Kilimanjaro Christian Medical Centre (KCMC). 

Consecutive patients presenting to the KCMC emergency department were screened and consenting patients with a Universal Vital Assessment (UVA) score of  ≥ 2 (moderate/high) were enrolled. Baseline demographic and clinical data were collected including AKI risk factors. In every patient, serum creatinine tests were done on admission and repeated every 24 - 48 hrs during hospital admission. AKI was determined using the KDIGO criteria based on a rise from the first creatinine or calculated baseline using the MDRD equation. Descriptive statistics were used to summarize participant characteristics. We used a modified Poisson regression analysis to identify factors independently associated with AKI. To validate the predictive performance of the variables in the AKI risk score, a regression model was generated and the Receiver Operating Curve (ROC) was plotted comparing the predicted probability against the observed AKI status. Model fit was assessed using Akaike Information Criterion (AIC). 

A total of 628 admissions were included in the analysis with a mean age of 58 (+/- 20.2) years. Most (75%) had a moderate UVA illness severity score. 35.5% (n= 223) met the KDIGO criteria for AKI over the study period. Amongst those with no AKI on admission and subsequent creatinine information (n=469), 25 (5.3%) developed a new episode of AKI during the hospital stay. Using the AKI risk variables, the collective model yielded an area under the ROC of 0.547 (95% CI: 0.499–0.594) for AKI prevalent cases and 0.51 (95% CI: 0.406–0.621) for new AKI cases indicating poor prediction of AKI (Figure 1). Diabetes and anemia were significantly associated with AKI in regression analysis (p < 0.001 and 0.01 respectively). In exploratory analysis, a model with comorbidities (hypertension, anaemia, diabetes) provided moderate prediction of AKI, area under ROC 0.65 (95% CI: 0.589–0.684). Overall, 27% of the patients died during the hospital admission (36% of those with AKI and 22% without AKI, p < 0.001). 

The proportion of patients with AKI on admission was very high in the emergency department in Northern Tanzania and associated with a significantly higher mortality. The selected AKI risk variables had poor discriminatory capacity for detecting AKI in this selected group of sick patients presenting to a tertiary hospital. Most AKI was community acquired and the lower incidence of hospital acquired AKI could reflect heightened awareness and resultant AKI prevention and treatment in at-risk patients in the centre. Further initiatives beyond 0by25 towards AKI could focus on community health centres.