Supplementary Data
Methods: Statistical Analysis of Risk Score
In a stepwise backward Cox regression analysis was carried out with all univariately associated
variables. After this model was constructed, optimal cut off points were determined by the
highest sum of specificity and sensitivity calculated using receiver operator curve analysis. Cut
offs for the model were then chosen at clinical relevant or numerical easy cut off points. In
sensitivity analysis, bootstrapped forward Cox regression was carried out via the swboot syntax
in STATA to investigate the relative importance of each included variable. Internal validation of
additive power of KIM-1 to the model was carried out by comparison of Harrell’s C indices in
two randomly created datasets from the entire study population by using the Somersd syntax in
STATA. With the dichotomized variables a new model was constructed, and based on this model
a risk score was created based on the individual X2 values of each variable in the multivariable
model. For simplicity, this model did not include interaction terms. Finally, quartiles of this risk
score were investigated for the relationship with WRF.
Results: Risk score for WRF
On the basis of the final stepwise Cox regression model, we determined optimal cut-off values
for each variable in the model. Table 3 shows the cut-off values and the multivariable HR for
each variable after dichotomization. Afterwards, we constructed a risk model for WRF with
weights of each variable based on Chi2 values as presented in the last column of Supplementary
Table 3.
The maximum score of this risk score is 54 points, which was present in 1 patient, while in 51
patients the minimum score of 0 was present. Mean risk score was 20 ± 11 points. Among
quartiles of the risk score, the risk for WRF increased with higher quartiles of the risk score.
However, quartile 2 and 3 did not differ in their risk for WRF. Therefore, for better risk
stratification, taking these groups together resulted in three groups (Supplementary Table 4). In
these three groups a risk score 13 – 28 was associated with increased risk for WRF HR 2.63,
95% CI 1.77 to 3.89, P < 0.001, while a risk score > 28 carried the highest risk: HR 6.08, 95%
CI 4.09 to 9.04, P < 0.001 (both versus risk < 13) (Supplementary Figure 1). Alternatively, since
the middle groups is a representation of the mean study population, the relative risk in relation to
a risk score of 13 – 28 for a risk score < 13 was HR 0.38, 95% CI 0.26 to 0.56, P < 0.001, and
for a risk score > 28 HR 2.32, 95% CI 1.80 to 2.96, P < 0.001.
Using the alternative definition of an absolute increase in serum creatinine (> 0.3mg/dL), we
found very similar results. A risk score 13 – 28 showed a HR 2.63, 95% CI 1.77 to 3.89, P < 0.001,
while a risk score > 28 carried also using this definition the highest risk for WRF: HR 7.56, 95% CI 5.21
to 11.0, P < 0.001. Bootstrapped Harrell’s C for this model with this definition of WRF was 0.703 (0.675
– 0.731), P < 0.001.
Supplementary Table 1. Variable selection on the basis of bootstrapped stepwise Cox
regression
Number of replications out of 100 in which factor selected
Variable
Model without
With KIM-1
With NAG
With NGAL
tubular markers
KIM-1
NA
85
NA
NA
NAG
NA
NA
4
NA
NGAL
NA
NA
NA
25
eGFR
79
72
81
79
UACR
89
30
85
79
BMI
58
44
45
56
LVEF
52
44
41
50
COPD
37
42
56
38
Furosemide use
38
34
33
39
ASA use
33
47
27
25
Hemoglobin
30
25
28
29
Age
46
21
34
39
Diabetes
20
21
13
16
Digoxin use
16
10
16
11
Creatinine
13
20
13
13
DBP
6
6
7
5
Nitrate use
11
14
6
11
CVP>6 cm H2O
5
15
13
8
Beta-blocker use
9
10
12
6
Amiodarone use
12
10
8
16
NYHA class III/IV
10
11
8
14
Spironolactone use
11
2
9
9
Heart Rate
7
5
5
5
Atrial Fibrillation
2
2
2
2
Gender
0
1
1
1
P value for entering and remaining in the model 0.01 and 0.02. Abbreviations as in Table 1.
Supplementary Table 2: Reclassification
Model with KIM-1
Model without KIM-1
Risk Reclassification
0% to < 5%
5% to < 10%
10% to <20%
> 20%
Total
Higher
Lower
Persons Included
54
2
0
0
56
No. of events
1
0
0
0
1
0
NA
No of nonevents
53
2
0
0
55
2
NA
Persons Included
56
452
100
0
608
No. of events
0
33
16
0
49
16
0
No of nonevents
56
419
84
0
559
84
56
Persons Included
0
153
656
161
879
No. of events
0
17
100
36
150
36
17
No of nonevents
0
136
556
125
817
125
136
0% to < 5%
5% to < 10%
10% to < 20%
> 20%
Persons Included
0
0
97
280
377
No. of events
0
0
10
77
87
NA
10
No of nonevents
0
0
87
203
290
NA
87
110
605
853
441
2011
1
50
126
113
290
52
27
109
555
727
318
1721
209
279
Total
Persons Included
No. of events
No of nonevents
Abbreviations: KIM-1: Kidney Injury Molecule 1
Supplementary Table 3. Risk Score for WRF and cut off points
Variable
Multivariable HR
p-Value
X2
Score
Factor
KIM (> 3500 ng/gCr)
1.87 (1.47 to 2.38)
< 0.001
25
15
1.7
eGFR (< 60 mL/min/1.73m2)
1.64 (1.28 to 2.09)
< 0.001
16
9
1.8
LVEF (< 30%)
1.45 (1.14 to 1.85)
0.002
9
6
1.5
COPD (yes)
1.50 (1.15 to 1.95)
0.003
9
6
1.5
Hb (< 14 g/dL)
1.46 (1.13 to 1.88)
0.004
8
5
1.6
Loop Diuretics (yes)
1.53 (1.11 to 2.09)
0.009
7
4
1.8
BMI (> 28 kg/m2)
1.34 (1.05 to 1.72)
0.021
5
3
1.7
Aspirin use (no)
1.31 (1.03 to 1.67)
0.027
5
3
1.7
Age (>75 years)
1.36 (1.04 to 1.77)
0.024
5
3
1.7
Total Score
54
Abbreviations as in Table 1. Factor is ratio between X2 and Score. Bootstrapped Harrell’s C
with 100 replications: 0.677 (0.646 to 0.708), p < 0.001) (total dataset). The risk score is
calculated based on the presence or absence of the variable specified in the first column. If
present, the number in the corresponding 4th column is added to the score.
Supplementary Table 4. Risk score for WRF
Variable
N
Hazard Ratio (95% CI)
P-Value
Risk Score 0 – 12
577
1.00 (ref)
-
Risk Score 13 – 19
473
2.64 (1.71 to 4.05)
< 0.001
Risk Score 20 – 28
512
2.61 (1.71 to 4.01)
< 0.001
Risk Score 29 – 54
449
6.08 (4.09 to 9.04)
< 0.001
Risk Score < 13
577
1.00 (ref)
-
Risk Score 13 – 28
985
2.63 (1.77 to 3.89)
< 0.001
Risk Score > 28
449
6.08 (4.09 to 9.04)
< 0.001
Risk Score < 13
577
0.38 (0.26 to 0.56)
< 0.001
Risk Score 13 – 28
985
1.00 (ref)
-
Risk Score > 28
449
2.32 (1.80 to 2.96)
< 0.001
These risk groups represent groups of patients with similar risk scores. These scores have
been calculated based on the calculation of the risk score in supplementary table 3.
Supplementary Figure 1. Risk Score and WRF