Appendix
Appendix Table A1. Acute kidney injury prediction score (APS)
Points Scored
0
Age (years)
<60
Respiratory Rate
<20
AVPU Score
Alert
1
2
3
60-79
≥80
≥20
Other
CKD Stage 3a-5
Y
Heart failure
Y
Y
Diabetes
Y
Liver disease
AVPU scale best response (alert, vocal, pain, unresponsive), CKD – chronic kidney disease
(eGFR <60mls/min), Respiratory rates - breaths/minute.
Appendix Table A2. Summary of key variable between validation (medicine with baseline
SCr) and original derivation study.
Validation
Derivation Study
Variable
HA-AKI
No HA-AKI
HA-AKI
No HA-AKI
Age
84 (76-89)
79 (67-86)
80 (70-86)
73 (61-81)
Diabetes
25%
21%
16%
8%
Heart failure
37%
18%
20%
8%
CKD
29%
10%
49%
28%
Liver disease
2%
1%
4%
2%
AVPU <Alert
1.7%
1%
4%
1%
RR ≥20
28%
23%
37%
22%
CKD – chronic kidney disease, HA-AKI – hospital-acquired AKI, RR – respiratory rate breaths
per minute.
Study,
population
External
validation
Country, Design, Outcome, AKI Definitions, Methods
Results, Variables, Model Performance
[31] Drawz
2008
Medicine,
surgery,
obstetrics
D, IV
TRIPOD 26
USA 3 centres, retrospective, case-control (n=180 cases, n=360 controls). 2003.
Outcome: In-hospital increase SCr ≥44μmol/L if base SCr ≤168μmol/L, ≥88μmol/L (base
177-433μmol/L) & ≥133μmol/L (base >442μmol/L). Admission SCr = baseline.
Controls: mix same discharge diagnosis or next patient admitted to clinical team.
Inclusions: age ≥18 & normal admission SCr or admission SCr not qualifying as AKI vs
known baseline. Exclusions: RRT, no repeat SCr.
19 Variables: demographics (age, sex, race), history, drugs, admission observations (BP,
pulse) & lab parameters (HCO3, urea, SCr, & albumin). Variables with p value <0.20
entered into multiple logistic regression model. Missing data excluded.
7 variables: Age, SBP, HR, HCO3, Urea, Albumin & drugs (NSAIDs,
ACE-I, ARBs, diuretic).
Derivation c-statistic 0.73.
Simplified model: HR ≥70/min, HCO3<24 or >30mmol/L), SCr ≥88μmol/L
& drugs. C-statistic derivation 0.69. IV both models C-Statistic 0.66.
Calibration: No H-L p value. Risk range in validation: 0-1: 16% vs 4 risk
factors: 62%. No information on mortality.
-
[32]
Matheny
2010
General
admissions
D
TRIPOD 27
USA single centre, retrospective cohort (n=61,179). 1999-2003.
Outcome (<30 days): AKI Risk = ≥2 SCr results ≥150% baseline. Injury = ≥200% base.
Inclusions: adult admissions ≥2 days.
Exclusions: missing data; baseline eGFR <60 (n=11,342), Admission AKI, no SCr <48 hrs
admission (n=10,378) or no repeat (n=13,352).
Included n=26,107. eGFR - MDRD.
27 variables: coded diagnoses (including at admission) & drugs following univariate
analysis placed in multivariable model. Missing values captured as separate category.
10-fold cross-validation employed to estimate overfitting.
27 variables: Female, Age, Race, 11 Drug classes, Contrast, Bacterial
infection (use antibiotics), mean SCr, MI, Rhabdomyolysis, Hepatitis,
Pancreatitis, Ammonia, AST/ALT ratio, Thrombocytopenia,
Leucocytosis, Hypercalcaemia, Glucose.
AKI Risk: AUROC 0.75 (0.73–0.76). Calibration: H-L P=0.29.
AKI Injury: AUROC 0.78 (0.76–0.79). H-L P=0.12.
AKI Risk 5.2% (n=1,352), AKI Injury 2.8% (n=726).
No mortality data.
-
[17] Forni
2013
General
medical
D, IV, EV
TRIPOD 29
UK single centre, prospectively collected data (n=3,707). 2011.
Outcome: AKI (KDIGO SCr change <7 days).
Exclusions: RRT, non-medical, age <18, admission AKI (n=184), missing data (n=553).
Included n=3,523. Derivation n=1,867. Pre-admission SCr >1 month & <6 months prior.
Internal validation: no previous SCr result, but with a normal SCr on admission (defined
80-120µmol/L) (n=1,656). CKD defined - eGFR <60.
25 variables on univariate, If P <0.05 variable entered into multivariable analysis. No
missing data information.
7 Variables: Age 60-79 (1pt) ≥80 (3pts), CCF, CKD, Diabetes (2pt),
Liver disease (3 pt), respiratory rate ≥20/min, <alert on AVPU (3 pts).
Derivation AUROC 0.72 (0.66–0.77). H-L P=0.96. Risks plotted.
Validation AUROC 0.76 (0.71–0.82). No H-L reported.
7% (n=95) incidence HA-AKI in Derivation with mortality 20% (n=19) vs
3.5% (n=62) without HA-AKI. In validation n=60 developed HA-AKI.
Current
study:
AUROC
0.65-0.71
[33]Bedford
2016
General
admissions
D, EV
TRIPOD 29
UK 3 centres, 1 used for EV (n=11,655). 2011.
Outcomes: AKI 72 hours (KDIGO SCr change) & grade 2/3 at 72 hours.
Exclusions: elective, admission AKI, childbirth, pregnancy, no info on AKI at 72 hours.
Used information gathered up to 24hrs after admission on 35 variables.
Ordinal logistic regression with univariable analysis prior to backwards selection for
retention of statistically significant variables for multivariate modelling. Missing data
excluded or given own category. 3:1 random split for IV. Derived in 7,556 admissions & IV
in 2,514 (n=6,626 individuals). EV: 1,585 admissions.
12 variables: Age, Primary diagnosis, Previous admissions, Charlson
co-morbidity score, HbA1C, Troponin, Proteinuria, eGFR, K+, WCC,
Mg2+, CRP.
AKI: AUROC 0.67 (0.64-0.71), H-L P=0.04. AKI stage 2/3 AUROC 0.68.
External validation: AKI AUROC 0.71, H-L P=0.12; AKI 2/3 AUROC 0.63,
H-L P=0.14.
Derivation incidence HA-AKI 8.9%. Internal validation: 9.6%, Stage 2/3:
1.6% ( n=40). External validation: AKI 7.6% (120/1585, Stage 2/3 0.7%
(n=12)
Same
study
AUROC
0.71
Appendix Table A3 - Prediction models for HA-AKI in general populations. ACE-I – angiotensin-converting enzyme inhibitors, ARBs – angiotensin 2 receptor blockers, AUROC – area under the
receiver operating characteristic curve, CKD – chronic kidney disease, D – derivation study, EV – external validation study, H-L – Hosmer-Lemeshow (goodness-of-fit test of calibration), IV - Internal
validation study, K+ - serum potassium, Mg2+ -serum magnesium, MDRD - Modification of Diet in Renal Disease equation, NSAIDs – non-steroidal anti-inflammatories, RRT – renal replacement
therapy, SCr – serum creatinine, TRIPOD score (out of 37) – recommended reporting points in prediction model studies, WCC – white cell count.