This study aimed to develop and validate an acute renal angina index (aRAI) to predict pediatric acute kidney injury (AKI) in patients presenting to the emergency department (ED) with possible sepsis. Researchers derived the aRAI based on variables available in the ED and tested its performance on 101 patients. The aRAI demonstrated good discrimination for predicting AKI, with an AUC of 0.92. Combining the aRAI with urinary neutrophil gelatinase-associated lipocalin further improved risk stratification. While promising, the study was limited by its single center design and heterogeneous population in the ED. Larger validation studies are still needed to evaluate the aRAI in broader pediatric populations.

1. S L I D E 0
Early prediction of pediatric acute kidney injury
from the emergency department: A pilot study
Journal Club

2. S L I D E 1

3. S L I D E 2
Reporting Standards and Critical Appraisal of Prediction Models
• Assess risk of bias
– Was an appropriate study design used to collect information for
model development?
– Was the target outcome in the development and validation cohorts
always defined the same way, objectively assessed in the same way and
were the outcomes assessors blinded to the values of the candidate
predictors?
– Was the number of candidate predictors and manipulation of the
predictors during statistical analysis (e.g. premature dichotomization
of continuous, categorical or ordinal values) reasonable for the number
of target events seen?
– Were missing values handled in an appropriate fashion?
– Was predictor selection and regression coefficient fitting performed
in a reasonable manner?
– Was the evaluation of model performance done in a sufficiently
independent dataset?
Wee et al. Standards and Critical Appraisal of Prediction Models. 2018

4. S L I D E 3
Reporting Standards and Critical Appraisal of Prediction Models
• Assess applicability
– Did the modelling study select a representative source of individual
data?
– Were there differences in the treatments administered (if any) that
does not match your question?
– Will the predictors, its definitions and its methods of measurement
match what you intend to do?
– Does the desired outcome, its definition and its method of
assessment match what you intend to do?
– Does the time point of the predicted event match what you intend to
use the model for?
– Is the performance of the model, in regards to calibration and
discrimination, fit for purpose in regards to the clinical decisions that
have to be made as a consequence of the prediction?
Wee et al. Standards and Critical Appraisal of Prediction Models. 2018

5. S L I D E 4
Background
• AKI affects 30% of children in ICU and 5% in non-ICU
• Identifying AKI early can inform medical decisions key to
mitigation of injury
• Currently, SCr is the standard method to diagnose AKI.
• SCr is a delayed and unreliable marker for AKI
• Renal angina index (RAI) has proven better than SCr alone at
predicting AKI in critically-ill children
• Hypothesis:
– Modifying the RAI to include “acute” components, available in
the ED, would help predict AKI in children with possible sepsis

6. S L I D E 5
Study objectives
A. To derive and test performance of an “acute” RAI (aRAI) in the
Emergency Department (ED) for prediction of inpatient AKI
B. To evaluate the added yield of urinary AKI biomarkers

7. S L I D E 6
Methods
• Study design and setting
– Prospective, observational cohort study
– Children >28 days to <25 years
– Single, high-volume, tertiary pediatric hospital
– 8/1/2015-5/9/2016
• Selection of participants
– Presented to the ED
– Concern for sepsis (sepsis alert)
– Admitted
– Had a SCr measured and urine specimen obtained
– Received fluid resuscitation
• Exclusion criteria
– Previously enrolled in this study
– CKD IV or V, anephric, or on dialysis.

8. S L I D E 7
Methods
• Data collection
– For each eligible subject, a manual chart review by 3 of the authors
– Variables: demographics, PMH, IVF, procedures, LOS, disposition, and
all measured SCr values
– Subjects were followed for 72 h (or until discharge if before)
– A left-over urine sample was obtained for biomarker testing
• Urinary AKI biomarkers
– Neutrophil gelatinase-associated lipocalin (NGAL)
– Kidney injury molecule-1 (KIM-1)
– Interleukin 18 (IL-18)
– Liver fatty acid binding protein (L-FABP)

9. S L I D E 8
Methods

10. S L I D E 9
Methods
• Outcomes
– Primary outcome -> KDIGO-defined AKI between 24 and 72 h
– Baseline SCr: lowest SCr in previous 6 mo
– If no baseline SCr -> CrCl 120 was imputed
• Sample size
– Assuming a 10% incidence of AKI in pediatric population with shock
and an area under the curve (AUC) for the RAI of 0.7–0.8
– To detect a difference in AUC by 0.10, and a two-sided test at 0.05
alpha, 116 subjects provided 80% power
– For a planned 116 total patients, expected 12 with AKI and 104 without

11. S L I D E 10
Statistical analysis
• Descriptive statistics for population characteristics
• RA(+) and RA(−) groups differences were assessed using
– Categorical -> chi-square or fisher's exact test
– Continuous variables -> t-tests for
• aRAI evaluated as a diagnostic test and compared to SCr
– Cut point of aRAI ≥8
– Sensitivity, specificity, PPV, NPV and ROC analysis
• Predict AKI -> simple and multivariable logistic regression
• Compare ROC-AUCs -> DeLong’s method
• Compare individual probability and risk of primary outcome ->
derivation of classification and regression tree analysis (separate
the entire cohort into terminal node cohorts)
• A p<0.05 was considered significant
• SAS and Stata used

12. S L I D E 11
Results

13. S L I D E 12
Results
(27%) (73%)

14. S L I D E 13
Results
AUC of 0.92 (0.86–0.98) for the prediction of AKI

15. S L I D E 14
Results

16. S L I D E 15
Results
The classification analysis identified a terminal node of RA(+)/NGAL(+) with a
probability of inpatient AKI of 60% higher than any of the other nodes
Biomarkers

17. S L I D E 16
83 subjects with a>1x increase
in SCr from baseline in the ED
35 subjects with a SCr <1x
baseline SCr in the ED
17 subjects developed AKI
between 24-72 h
101 subjects with no inpt AKI
(n=76) or no repeat SCr (n=25)
16 RA(+)

18. S L I D E 17
Limitations
• Observational study -> causation cannot be assessed
• Chart review -> risks of inaccurate data
• Pilot study at a single center -> may not be generalizable
• Sick population (possible sepsis) -> may not be generalizable to a
heterogeneous population
• Imputation of SCr using the Schwartz formula
• Not all subjects had a daily SCr -> incomplete information

19. S L I D E 18
Conclusion
• The aRAI was shown to be a sensitive test that can be used in the
ED and that outperforms using a change in SCr to predict AKI after
admission to the hospital
• In the future, this tool should be evaluated in a broadened,
heterogenous population

20. S L I D E 19
Reporting Standards and Critical Appraisal of Prediction Models
• Assess risk of bias
– Was an appropriate study design used to collect information for
model development?
– Was the target outcome in the development and validation cohorts
always defined the same way, objectively assessed in the same way and
were the outcomes assessors blinded to the values of the candidate
predictors?
– Was the number of candidate predictors and manipulation of the
predictors during statistical analysis (e.g. premature dichotomization
of continuous, categorical or ordinal values) reasonable for the number
of target events seen?
– Were missing values handled in an appropriate fashion?
– Was predictor selection and regression coefficient fitting performed
in a reasonable manner?
– Was the evaluation of model performance done in a sufficiently
independent dataset?
Wee et al. Standards and Critical Appraisal of Prediction Models. 2018

21. S L I D E 20
Reporting Standards and Critical Appraisal of Prediction Models
• Assess applicability
– Did the modelling study select a representative source of individual
data?
– Were there differences in the treatments administered (if any) that
does not match your question?
– Will the predictors, its definitions and its methods of measurement
match what you intend to do?
– Does the desired outcome, its definition and its method of
assessment match what you intend to do?
– Does the time point of the predicted event match what you intend to
use the model for?
– Is the performance of the model, in regards to calibration and
discrimination, fit for purpose in regards to the clinical decisions that
have to be made as a consequence of the prediction?
Wee et al. Standards and Critical Appraisal of Prediction Models. 2018

22. S L I D E 21
Thank you

23. S L I D E 22
Extra Slides

24. S L I D E 23
KDIGO AKI Definition
Go back

25. S L I D E 24
AKI biomarkers
Go back