APPENDIX
CONFIRM – optimized CCTA score
Modeling details
1. Internal Validation:
To avoid overfitting during modeling of the score the test sample was randomly
divided into two equally sized groups, of which one was used for modeling and the
other for internal validation of the models. Randomization was stratified by outcome
(death vs. no death) to ensure the same number of events in both groups
2. Testing of candidate parameters.
The predictive value of candidate parameters was tested using Cox-proportional
hazard model. Analysis was corrected for pre-test risk assessed by NCEP ATP III risk
score. Performance was assessed by chi2 and corrected C-index.
2.1. Simple parameters and published scores
Stenosis
Number of segments with stenosis >0%
Number of segments with stenosis >50%
Number of segments with stenosis >70%
Number of vessels with stenosis >0%
Number of vessels with stenosis >50%
Number of vessels with stenosis >70%
Plaque
Number of segments with any plaque
Number of segments with noncalcified plaques
Number of segments with mixed plaques
Number of segments with calcified plaques
Number of segments with mixed or calcified plaques
Published scores
CAD involvement (Chow)
Segment Stenosis score (Min)
Duke CAD
Duke Jeopardy
Coefficient
chi²
C-index
0.104
0.191
0.283
0.345
0.361
0.502
19.6
25.8
40.9
27.0
25.5
32.5
0.57
0.58
0.56
0.61
0.58
0.56
0.113
0.113
0.108
0.124
22.4
0.47
12.0
12.9
26.6
0.59
0.50
0.55
0.55
0.61
0.316
0.064
0.218
0.311
27.6
35.2
38.7
43.4
0.60
0.60
0.60
0.62
2.2. Segment jeopardy
To explore the effect on lesion localization a segment specific univariable analysis
was performed:
chi²
Segment
Proximal RCA
Mid RCA
Distal RCA
R. posterolateralis dexter
Left main coronary artery
Proximal LAD
Mid LAD
Distal LAD
R diagonalis 1
R diagonalis 2
Proximal circumflex
Obtuse marginal branch 1
Distal circumflex
Obtuse marginal branch 2
R. posterolateralis sinister
Stenosis>0%
Stenosis>50%
9,29
8,94
4,32
4,81
15,07
14,17
0,57
1,22
3,23
0,02
15,31
9,10
3,85
3,63
0,05
0,87
14,70
1,17
0,86
0,57
25,44
4,20
1,25
5,14
0,25
14,81
10,02
4,84
8,82
0,01
Calcified or mixed
plaque
19,12
10,29
3,81
2,55
16,53
29,12
1,46
0,25
6,70
0,72
12,18
14,36
6,66
6,07
7,84
proximal RCA, mid RCA, left main, proximal LAD, mid LAD, proximal circumflex,
and obtuse marginal branch 1) into analysis improved the predictive value slightly but
significantly. Mid LAD was included because of the difficulty of an exact
differentiation from the proximal LAD
Number of proximal segments with stenosis >50%
Number of proximal segments with mixed or calcified plaques
Chi2
24.7
24.8
C-index
0.53
0.60
After inclusion of these 7 Segments, the other segments lost significance in all 3
models.
The effect of higher numbers of proximal segments affected was limited and had a
high variance. Further improvement of prediction could be reached by only counting
the first two segments with plaque and the first two segments with stenosis:
Number of proximal segments with stenosis >50% (2 at most)
Number of prox. segments with mixed or calcified plaques(2 at most)
Chi2
35.1
33.7
C-index
0.55
0.62
3.2. creating the optimized score
The optimized score should consist of 3 parts, for which the following candidate
parameters were tested:
1. clinical risk
NCEP ATP III
2. plaque composition
Number of vessels with any plaque
Number of vessels with calcified plaque
Number of proximal/any segments with any plaque
Number of proximal/any segments with calcified plaque
3. stenosis
Number of vessels with stenosis >50%
Number of proximal/any segments with stenosis >50%
The final model consisted of these 3 parameters:
NCEP ATP III
Number of proximal segments with stenosis >50% (2 at most)
Number of prox. segments with mixed or calcified plaques (2 at most)
Coefficient
0.144
0.398
0.407
Chi2
28.1
15.1
13.1
NCEP ATP III was used in the form of ln(NCEP 10 year risk) /0.235 to model 1 score
point in the NCEP ATP III point model. Thus, each of the CCTA parameters is roughly
equivalent to 2.8 NCEP ATP III score points. The optimized score was calculated
according to the formula:
1 ln(NCEP 10 year risk) /0.235 +
2.83 number of proximal segments with calcified or mixed plaques (2 at most) +
2.76 number of proximal segments with stenosis >50% (2 at most)
3.3 calibration
1
2
3
4
5
y = 0.1207514 * exp(0.1324839 * x)
0
Annual mortality rate in percent
6
Calculation of mortality risk was done using the cox proportional hazard model and
the base hazard of the test sample. For a 2 year observation period the base hazard for
a score of zero was 0.1207514 (expressed as annual mortality in percentage). The
weighting coefficient for the optimized score was 0.1324839. The calibration curve is
plotted below:
-10
0
10
20
30
Points of the optimized score
The endpoint of this study (total mortality) is different from the endpoint in NCEP
ATP III (cardiac death or myocardial infarction). The annual mortality rate in the test
sample was 0.75%, while the predicted annual rate for hard cardiac events according
to NCEP ATP III in the same sample was 0.95%. To ascertain compatibility with
existing risk scores the cutoff values between low and intermediate risk and between
intermediate and high risk were therefore corrected by the quotient of 0.8. The risk
groups were therefore defined as:
Low risk:
Intermediate risk:
High risk
<0.8% per year
between 0.8 and 1.6& per year
>1.6% per year.
4. Validation of the score
4.1. Internal validation
For validation z-values in C- and IDI-Statistics were calculated for improvement over
NCEP ATP III risk score in the test and validation sample
Internal test sample
Internal validation sample
Bootstrap mean
Bootstrap standard dev.
z C-statistics
6.33
6.11
5.67
0.87
z NRI statistics
6.43
6.41
5.62
0.98
z C-statistics
2.06
z NRI statistics
3.26
4.2 External validation
Total sample
4.3. Subgroup analysis
The performance of the score was tested on clinically relevant subgroups of patients
and on the participating sites. Analysis was only done, if more than 5 events were
recorded in the group of interest. Hazard ratio is calculated between the 75th percentile
and the 25th percentile of the optimized score within the group of interest.
Hazard Ratio
0.5
1
All patients
Age <50
50-65
>65
Sex female
male
NCEP ATP III low
2
Hazard Ratio
4
6
8
0.5
Test sites
Site 1
2
3
4
5
6
7
intermediate
8
high
9
obstructive CAD no
yes
Validation sites
Site 1
2
1
2
4
8
16
24